diff --git a/.goreleaser.yaml b/.goreleaser.yaml
index af2e53c..984b2a2 100644
--- a/.goreleaser.yaml
+++ b/.goreleaser.yaml
@@ -30,8 +30,8 @@ builds:
       # Commit is what identifies a dev build: they all share one moving
       # release tag, so the version string can't tell two of them apart.
       - -s -w
-        -X github.com/coder/vllm-helper/internal/cli.Version={{.Version}}
-        -X github.com/coder/vllm-helper/internal/cli.Commit={{.FullCommit}}
+        -X github.com/coder/llm-helper/internal/cli.Version={{.Version}}
+        -X github.com/coder/llm-helper/internal/cli.Commit={{.FullCommit}}
   # The same binary under its old name. 'upgrade' in v0.5.0 and earlier
   # unpacks the file called vllm-helper and nothing else, so dropping it would
   # strand every existing install on the release it has.
diff --git a/AGENTS.md b/AGENTS.md
index 690b3ef..048c4de 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -23,7 +23,7 @@ llm-helper is one Go binary that manages inference servers (vLLM, mlx-lm, llama.
 
 ## Repository layout
 
-`internal/` holds the packages the binary builds from: `cli` for the commands, `control` for what the CLI and the web panel share, `web` for the panel, and the rest one package per feature. `catalogue/` holds the shipped data definitions (engines, recipes, hardware) with the Go that embeds them in `catalogue/catalogue.go`. `docs/` is the shipped documentation, and `tools/recipediff` is a separate module.
+`internal/` holds the packages the binary builds from: `cli` for the commands, `control` for what the CLI and the web panel share, `web` for the panel, and the rest one package per feature. `catalogue/` holds the shipped data definitions (engines, recipes, hardware) with the Go that embeds them in `catalogue/catalogue.go`; each kind is read by the `internal/` package of the same name (`engine`, `recipe`, `hardware`), and `resolve` combines them with a model's card and config into launch settings. `docs/` is the shipped documentation, and `tools/recipediff` is a separate module.
 
 ## Essential commands
 
diff --git a/catalogue/engines/README.md b/catalogue/engines/README.md
index 049c732..8e5ff01 100644
--- a/catalogue/engines/README.md
+++ b/catalogue/engines/README.md
@@ -5,7 +5,7 @@ Each file here is one inference server llm-helper can launch, named after the en
 
 To add an engine, copy the closest file to `<id>.json` and change it.
 No Go code is needed: shared code acts on the engine's `capabilities`, never on its name.
-The file is checked when it loads, and the tests in `internal/provider` render every file here for every rank.
+The file is checked when it loads, and the tests in `internal/engine` render every file here for every rank.
 
 The fields are described in "Contributing an engine" in [docs/providers.md](../../docs/providers.md).
 To try an engine without rebuilding, put its file in `~/.config/vllm-helper/engines.d/` instead.
diff --git a/cmd/vllm-helper-docs/main.go b/cmd/vllm-helper-docs/main.go
index f86cc3b..701bef4 100644
--- a/cmd/vllm-helper-docs/main.go
+++ b/cmd/vllm-helper-docs/main.go
@@ -22,7 +22,7 @@ import (
 	"syscall"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/web"
+	"github.com/coder/llm-helper/internal/web"
 )
 
 func main() {
diff --git a/docs/docs_test.go b/docs/docs_test.go
index f1973d6..e741e8f 100644
--- a/docs/docs_test.go
+++ b/docs/docs_test.go
@@ -6,7 +6,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 // The manifest is the docs engine's contract with the panel: one entry per
diff --git a/docs/providers.md b/docs/providers.md
index 67444f0..ad22ab2 100644
--- a/docs/providers.md
+++ b/docs/providers.md
@@ -185,7 +185,7 @@ A definition holds `schema_version` and the engine's fields, and says:
 The definition is checked when it loads.
 An unknown field, an unknown placeholder, a key that would land on the command line, or a container with no role that serves HTTP is refused with the reason.
 A file whose name is not its id is refused too.
-The tests in `internal/provider` render every shipped engine for every rank, so a new engine is covered once its file is in `catalogue/engines/`.
+The tests in `internal/engine` render every shipped engine for every rank, so a new engine is covered once its file is in `catalogue/engines/`.
 
 To try an engine without rebuilding, put its file under `~/.config/vllm-helper/engines.d/`, in the same shape and named the same way as a file in `catalogue/engines/`.
 A local file can only add engines with new ids; it cannot change how a shipped engine is launched.
diff --git a/go.mod b/go.mod
index c12f86f..bf5b0ab 100644
--- a/go.mod
+++ b/go.mod
@@ -1,4 +1,4 @@
-module github.com/coder/vllm-helper
+module github.com/coder/llm-helper
 
 go 1.26.5
 
diff --git a/internal/cli/capabilities.go b/internal/cli/capabilities.go
index 4de2172..ae80d27 100644
--- a/internal/cli/capabilities.go
+++ b/internal/cli/capabilities.go
@@ -4,8 +4,8 @@ import (
 	"fmt"
 	"os"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // vllmCaps asks the configured vLLM what it supports, caching under the state
diff --git a/internal/cli/clusterrunner.go b/internal/cli/clusterrunner.go
index 127237e..7c0bf17 100644
--- a/internal/cli/clusterrunner.go
+++ b/internal/cli/clusterrunner.go
@@ -10,11 +10,11 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/recipe"
-	"github.com/coder/vllm-helper/internal/service"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/service"
 )
 
 // clusterRunner drives a recipe that serves from containers on the nodes it
diff --git a/internal/cli/clusterrunner_internal_test.go b/internal/cli/clusterrunner_internal_test.go
index 17c302a..296401a 100644
--- a/internal/cli/clusterrunner_internal_test.go
+++ b/internal/cli/clusterrunner_internal_test.go
@@ -5,7 +5,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/nodes"
 )
 
 // status reads this vocabulary, so a mistranslation here is what makes a
diff --git a/internal/cli/detectcmds.go b/internal/cli/detectcmds.go
index 5100d3d..2b52c98 100644
--- a/internal/cli/detectcmds.go
+++ b/internal/cli/detectcmds.go
@@ -10,16 +10,16 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
-	"github.com/coder/vllm-helper/internal/toolchain"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/toolchain"
 )
 
 // doctorReport is everything doctor learns, in one machine-readable shape.
@@ -139,14 +139,14 @@ func gatherDoctor() doctorReport {
 			HarmonyVocab:    hasHarmonyVocab(cfg.Dirs.Data),
 			CoderConfigured: cfg.CoderConfigured(),
 		}
-		if prov, err := provider.Lookup(cfg.Provider()); err != nil {
+		if prov, err := engine.Lookup(cfg.Provider()); err != nil {
 			r.Config.Provider = cfg.Provider()
 			r.Config.ProviderError = err.Error()
 		} else {
 			r.Config.Provider = prov.Name()
-			if prov.Name() != provider.VLLM {
+			if prov.Name() != engine.VLLM {
 				r.Config.ProviderBin = cfg.Lookup(prov.BinKey(), "")
-				r.Config.ProviderBinOK = provider.Executable(r.Config.ProviderBin)
+				r.Config.ProviderBinOK = engine.Executable(r.Config.ProviderBin)
 			}
 		}
 	}
@@ -274,7 +274,7 @@ func hardwareWarnings(hw detect.Hardware) []string {
 	for _, err := range hardware.LocalErrors() {
 		w = append(w, "a local hardware definition was skipped: "+err.Error())
 	}
-	for _, err := range provider.LocalErrors() {
+	for _, err := range engine.LocalErrors() {
 		w = append(w, "a local engine definition was skipped: "+err.Error())
 	}
 	for _, err := range recipe.LocalErrors() {
@@ -387,7 +387,7 @@ func renderDoctor(r doctorReport, yes bool) {
 		if hw.GPUsHidden > 0 {
 			fmt.Printf("              %d more hidden from the server by CUDA_VISIBLE_DEVICES\n", hw.GPUsHidden)
 		}
-		newProfiles := provider.VLLM
+		newProfiles := engine.VLLM
 		if r.Config != nil {
 			newProfiles = r.Config.Provider
 		}
@@ -433,8 +433,8 @@ func renderDoctor(r doctorReport, yes bool) {
 		switch {
 		case c.ProviderError != "":
 			fmt.Printf("  provider:   %s\n", c.ProviderError)
-		case c.Provider != provider.VLLM:
-			p, _ := provider.Lookup(c.Provider)
+		case c.Provider != engine.VLLM:
+			p, _ := engine.Lookup(c.Provider)
 			fmt.Printf("  provider:   %s (new profiles)\n", c.Provider)
 			// "server", not "<provider> bin": the line above names the provider,
 			// and a longer name would push the value out of the column.
@@ -451,7 +451,7 @@ func renderDoctor(r doctorReport, yes bool) {
 			}
 		}
 		switch {
-		case c.VLLMBin == "" && c.Provider != provider.VLLM:
+		case c.VLLMBin == "" && c.Provider != engine.VLLM:
 			// Not a fault: vLLM is optional once new profiles go elsewhere.
 		case c.VLLMBin == "":
 			fmt.Println("  vllm bin:   NOT SET -- set VLLM_BIN in config.env")
@@ -789,7 +789,7 @@ func adoptService(cfg *config.Config, s detect.Server) bool {
 // server is not vLLM's backend, and telling its user to install vllm-metal
 // would send them after something they do not need.
 func appleBackend(hw detect.Hardware, newProfiles string) string {
-	if hw.GPUVendor != "apple" || newProfiles != provider.VLLM {
+	if hw.GPUVendor != "apple" || newProfiles != engine.VLLM {
 		return ""
 	}
 	if !hw.Metal {
diff --git a/internal/cli/doctor_internal_test.go b/internal/cli/doctor_internal_test.go
index f6fccb2..5a2eb69 100644
--- a/internal/cli/doctor_internal_test.go
+++ b/internal/cli/doctor_internal_test.go
@@ -13,10 +13,10 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // Nothing in this package's tests runs in parallel: they swap os.Stdout and
diff --git a/internal/cli/editset.go b/internal/cli/editset.go
index 988a2d1..7f4b6e4 100644
--- a/internal/cli/editset.go
+++ b/internal/cli/editset.go
@@ -10,9 +10,9 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // standardProfileKeys are the keys every profile is expected to carry.
@@ -24,7 +24,7 @@ var standardProfileKeys = map[string]bool{
 	"KV_CACHE_MEMORY_BYTES":  true,
 	"EXTRA_ARGS":             true,
 	profile.RecipeKey:        true,
-	provider.Key:             true,
+	engine.Key:               true,
 }
 
 func newEditCmd() *cobra.Command {
@@ -164,7 +164,7 @@ func validateProfile(p *profile.Profile, name string) error {
 	if p.Model() == "" {
 		return fmt.Errorf("profile '%s' has no MODEL= line -- it can't be served like this", name)
 	}
-	if _, err := provider.Lookup(p.Provider()); err != nil {
+	if _, err := engine.Lookup(p.Provider()); err != nil {
 		return fmt.Errorf("profile '%s': %w", name, err)
 	}
 	if extra := p.ExtraArgs(); extra != "" {
diff --git a/internal/cli/engines_internal_test.go b/internal/cli/engines_internal_test.go
index f7ce624..c33982f 100644
--- a/internal/cli/engines_internal_test.go
+++ b/internal/cli/engines_internal_test.go
@@ -6,9 +6,9 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 func glmRecipe(t *testing.T) recipe.TestedRecipe {
@@ -63,9 +63,9 @@ func TestCheckStagedNamesWhatIsMissing(t *testing.T) {
 		t.Fatalf("%s pins %d models into cache %q", r.ID, len(refs), cache)
 	}
 	weights, drafter := refs["weights"], refs["drafter"]
-	stage := func(m provider.ModelRef) {
+	stage := func(m engine.ModelRef) {
 		t.Helper()
-		dir := filepath.Join(home, provider.SnapshotPath(cache, m))
+		dir := filepath.Join(home, engine.SnapshotPath(cache, m))
 		if err := os.MkdirAll(dir, 0o755); err != nil {
 			t.Fatal(err)
 		}
@@ -92,7 +92,7 @@ func TestCheckStagedNamesWhatIsMissing(t *testing.T) {
 
 	// A snapshot directory with nothing in it is a download that did not
 	// finish.
-	if err := os.MkdirAll(filepath.Join(home, provider.SnapshotPath(cache, drafter)), 0o755); err != nil {
+	if err := os.MkdirAll(filepath.Join(home, engine.SnapshotPath(cache, drafter)), 0o755); err != nil {
 		t.Fatal(err)
 	}
 	if err := checkStaged(r, []recipeMember{local}, ""); err == nil || !strings.Contains(err.Error(), drafter.ID) {
diff --git a/internal/cli/engines_test.go b/internal/cli/engines_test.go
index 60e0196..76ad76e 100644
--- a/internal/cli/engines_test.go
+++ b/internal/cli/engines_test.go
@@ -7,8 +7,8 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 const toyEngine = `{"schema_version":1,"id":"toy","display_name":"Toy","why":"a test engine",` +
@@ -21,7 +21,7 @@ const toyEngine = `{"schema_version":1,"id":"toy","display_name":"Toy","why":"a
 // named rather than taking the others with it.
 func TestRunEnginesListShowsLocalDefinitions(t *testing.T) {
 	configDir, _ := newTree(t)
-	dir := filepath.Join(configDir, provider.LocalDir)
+	dir := filepath.Join(configDir, engine.LocalDir)
 	if err := os.MkdirAll(dir, 0o755); err != nil {
 		t.Fatal(err)
 	}
@@ -37,7 +37,7 @@ func TestRunEnginesListShowsLocalDefinitions(t *testing.T) {
 	}
 	// Later tests in this process must not see this directory's engines.
 	t.Cleanup(func() {
-		provider.UseLocal(filepath.Join(dir, "gone"))
+		engine.UseLocal(filepath.Join(dir, "gone"))
 		recipe.UseLocal(filepath.Join(dir, "gone"))
 	})
 
diff --git a/internal/cli/enginescmd.go b/internal/cli/enginescmd.go
index c9823e0..bfe16d3 100644
--- a/internal/cli/enginescmd.go
+++ b/internal/cli/enginescmd.go
@@ -10,8 +10,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
 )
 
 func newEnginesCmd() *cobra.Command {
@@ -62,12 +62,12 @@ func newEnginesShowCmd() *cobra.Command {
 
 // engineRow is one engine as 'engines list --json' prints it.
 type engineRow struct {
-	provider.Engine
+	engine.Engine
 	// Source is the engines.d file a local engine came from, or "shipped".
 	Source string `json:"source"`
 }
 
-func engineSource(e provider.Engine) string {
+func engineSource(e engine.Engine) string {
 	if e.Source == "" {
 		return "shipped"
 	}
@@ -75,11 +75,11 @@ func engineSource(e provider.Engine) string {
 }
 
 func runEnginesList(out, errOut io.Writer, jsonOut bool) error {
-	for _, err := range provider.LocalErrors() {
+	for _, err := range engine.LocalErrors() {
 		fmt.Fprintf(errOut, "warning: skipped a local engine definition file: %v\n", err)
 	}
 	var rows []engineRow
-	for _, e := range provider.Default().Engines() {
+	for _, e := range engine.Default().Engines() {
 		rows = append(rows, engineRow{Engine: e, Source: engineSource(e)})
 	}
 	if jsonOut {
@@ -93,14 +93,14 @@ func runEnginesList(out, errOut io.Writer, jsonOut bool) error {
 	if err := tw.Flush(); err != nil {
 		return err
 	}
-	fmt.Fprintf(out, "\nLocal engines go in %s.\n", filepath.Join(config.ResolveDirs().Config, provider.LocalDir))
+	fmt.Fprintf(out, "\nLocal engines go in %s.\n", filepath.Join(config.ResolveDirs().Config, engine.LocalDir))
 	return nil
 }
 
 func runEnginesShow(out io.Writer, id string, jsonOut bool) error {
-	e, ok := provider.Default().Get(id)
+	e, ok := engine.Default().Get(id)
 	if !ok {
-		return fmt.Errorf("no engine %q (have: %s)", id, strings.Join(provider.Names(), ", "))
+		return fmt.Errorf("no engine %q (have: %s)", id, strings.Join(engine.Names(), ", "))
 	}
 	if jsonOut {
 		return printJSONTo(out, engineRow{Engine: e, Source: engineSource(e)})
diff --git a/internal/cli/gendocs.go b/internal/cli/gendocs.go
index 305a27e..39d3c61 100644
--- a/internal/cli/gendocs.go
+++ b/internal/cli/gendocs.go
@@ -15,7 +15,7 @@ import (
 	"github.com/spf13/cobra"
 	"github.com/spf13/pflag"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 // CLIReferencePath is the manifest route whose children GenerateCLIRoutes
diff --git a/internal/cli/gendocs_internal_test.go b/internal/cli/gendocs_internal_test.go
index cb75425..e31703a 100644
--- a/internal/cli/gendocs_internal_test.go
+++ b/internal/cli/gendocs_internal_test.go
@@ -8,7 +8,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 // The committed reference must be exactly what the current command tree
diff --git a/internal/cli/ggufplugin.go b/internal/cli/ggufplugin.go
index a9eb6f7..a2cb2a7 100644
--- a/internal/cli/ggufplugin.go
+++ b/internal/cli/ggufplugin.go
@@ -8,8 +8,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
 )
 
 // ggufMissingMessage explains the gap and how to close it, naming the
diff --git a/internal/cli/hardware_test.go b/internal/cli/hardware_test.go
index b8433e9..dc4035c 100644
--- a/internal/cli/hardware_test.go
+++ b/internal/cli/hardware_test.go
@@ -7,7 +7,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // A hardware.d file is loaded before every command, so 'hardware list' shows
diff --git a/internal/cli/hardwarecmd.go b/internal/cli/hardwarecmd.go
index 22f6d4b..b165c8c 100644
--- a/internal/cli/hardwarecmd.go
+++ b/internal/cli/hardwarecmd.go
@@ -9,9 +9,9 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 func newHardwareCmd() *cobra.Command {
diff --git a/internal/cli/memfit.go b/internal/cli/memfit.go
index 6de1675..2574ae8 100644
--- a/internal/cli/memfit.go
+++ b/internal/cli/memfit.go
@@ -5,9 +5,9 @@ import (
 	"math"
 	"strconv"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // useServerGPUs tells detect which GPUs a server started with these env
diff --git a/internal/cli/memfit_darwin.go b/internal/cli/memfit_darwin.go
index c9837de..f69f122 100644
--- a/internal/cli/memfit_darwin.go
+++ b/internal/cli/memfit_darwin.go
@@ -3,7 +3,7 @@ package cli
 import (
 	"fmt"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // fitMemUtilization keeps --gpu-memory-utilization within what this Mac can
diff --git a/internal/cli/memfit_darwin_internal_test.go b/internal/cli/memfit_darwin_internal_test.go
index c9b1275..6e6a4ba 100644
--- a/internal/cli/memfit_darwin_internal_test.go
+++ b/internal/cli/memfit_darwin_internal_test.go
@@ -5,7 +5,7 @@ package cli
 import (
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // vllm-metal sizes its cache against the Metal working set and checks for
diff --git a/internal/cli/memfit_internal_test.go b/internal/cli/memfit_internal_test.go
index 95291b6..2cbae8b 100644
--- a/internal/cli/memfit_internal_test.go
+++ b/internal/cli/memfit_internal_test.go
@@ -4,7 +4,7 @@ import (
 	"strconv"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 func TestClampMemUtilization(t *testing.T) {
diff --git a/internal/cli/memfit_linux.go b/internal/cli/memfit_linux.go
index dbf5a18..54f7f02 100644
--- a/internal/cli/memfit_linux.go
+++ b/internal/cli/memfit_linux.go
@@ -3,7 +3,7 @@ package cli
 import (
 	"fmt"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // fitMemUtilization keeps --gpu-memory-utilization within what this machine can
diff --git a/internal/cli/metal_darwin.go b/internal/cli/metal_darwin.go
index f73ac80..a4ceb14 100644
--- a/internal/cli/metal_darwin.go
+++ b/internal/cli/metal_darwin.go
@@ -1,8 +1,8 @@
 package cli
 
 import (
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // useMetalProbe tells detect how to find out whether bin serves on the Apple
diff --git a/internal/cli/metallimit.go b/internal/cli/metallimit.go
index d3d9394..14b4e0a 100644
--- a/internal/cli/metallimit.go
+++ b/internal/cli/metallimit.go
@@ -11,7 +11,7 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // metalWiredLimitKey is the config.env key that records the wired limit
diff --git a/internal/cli/metallimit_darwin.go b/internal/cli/metallimit_darwin.go
index 6014bf1..f9a3064 100644
--- a/internal/cli/metallimit_darwin.go
+++ b/internal/cli/metallimit_darwin.go
@@ -9,9 +9,9 @@ import (
 	"strings"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 func runMetalLimit(o metalLimitOpts) error {
diff --git a/internal/cli/metallimit_internal_test.go b/internal/cli/metallimit_internal_test.go
index f894fd8..e249d6a 100644
--- a/internal/cli/metallimit_internal_test.go
+++ b/internal/cli/metallimit_internal_test.go
@@ -11,7 +11,7 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 const (
diff --git a/internal/cli/metallimit_other.go b/internal/cli/metallimit_other.go
index 791621d..b90df58 100644
--- a/internal/cli/metallimit_other.go
+++ b/internal/cli/metallimit_other.go
@@ -5,8 +5,8 @@ package cli
 import (
 	"errors"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // runMetalLimit refuses off macOS. The command still exists here so the
diff --git a/internal/cli/node_test.go b/internal/cli/node_test.go
index 3940298..e2a6119 100644
--- a/internal/cli/node_test.go
+++ b/internal/cli/node_test.go
@@ -4,7 +4,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/nodes"
 )
 
 // Re-adding a node is how its URL gets corrected. It used to replace the
diff --git a/internal/cli/nodecmd.go b/internal/cli/nodecmd.go
index 44c74eb..b332378 100644
--- a/internal/cli/nodecmd.go
+++ b/internal/cli/nodecmd.go
@@ -14,8 +14,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/nodes"
 )
 
 func newNodeCmd() *cobra.Command {
diff --git a/internal/cli/nodecmd_internal_test.go b/internal/cli/nodecmd_internal_test.go
index be77fa1..0099d4a 100644
--- a/internal/cli/nodecmd_internal_test.go
+++ b/internal/cli/nodecmd_internal_test.go
@@ -7,9 +7,9 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/nodes"
 )
 
 // 'node ls' showed only the HTTP host, so a node whose SSH target was wrong,
diff --git a/internal/cli/plugins.go b/internal/cli/plugins.go
index e0ed284..b3b7915 100644
--- a/internal/cli/plugins.go
+++ b/internal/cli/plugins.go
@@ -3,13 +3,13 @@ package cli
 import (
 	"fmt"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 
 	// Integrations register themselves. Importing them here, rather than
 	// from the plugin package, keeps the dependency pointing one way:
 	// plugins may use the core, the core never imports a plugin.
-	_ "github.com/coder/vllm-helper/internal/plugin/coder"
-	_ "github.com/coder/vllm-helper/internal/plugin/spark"
+	_ "github.com/coder/llm-helper/internal/plugin/coder"
+	_ "github.com/coder/llm-helper/internal/plugin/spark"
 )
 
 // localBaseURL is where this machine serves the OpenAI-compatible API.
diff --git a/internal/cli/poolserve.go b/internal/cli/poolserve.go
index f76501b..e19cc5d 100644
--- a/internal/cli/poolserve.go
+++ b/internal/cli/poolserve.go
@@ -15,8 +15,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/router"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/router"
 )
 
 func newPoolServeCmd() *cobra.Command {
diff --git a/internal/cli/profiles.go b/internal/cli/profiles.go
index 9e38e21..4e1063c 100644
--- a/internal/cli/profiles.go
+++ b/internal/cli/profiles.go
@@ -10,13 +10,13 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/catalog"
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/resolve"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // profileInfo is the machine-readable shape of one saved profile.
@@ -195,11 +195,11 @@ func newNewCmd() *cobra.Command {
 // Refusing here rather than at startup matters: activating a profile restarts
 // the service, so a model that cannot load takes the running one down with it
 // and leaves the machine serving nothing.
-func resolveQuant(ctx context.Context, cfg *config.Config, repo, want string) (*catalog.Choice, error) {
+func resolveQuant(ctx context.Context, cfg *config.Config, repo, want string) (*resolve.Choice, error) {
 	if want == "" {
 		return nil, nil
 	}
-	prov, err := provider.Lookup(cfg.Provider())
+	prov, err := engine.Lookup(cfg.Provider())
 	if err != nil {
 		return nil, err
 	}
@@ -207,14 +207,14 @@ func resolveQuant(ctx context.Context, cfg *config.Config, repo, want string) (*
 		return nil, fmt.Errorf("--quant picks one GGUF file out of a repo, but new profiles here are served by %s, which does not load GGUF; drop --quant and name a repo of that quant instead", prov.Name())
 	}
 	if !prov.Capabilities().VLLMArgs {
-		return catalog.ChooseGGUFQuant(ctx, repo, want)
+		return resolve.ChooseGGUFQuant(ctx, repo, want)
 	}
 	m, err := hf.Lookup(ctx, repo)
 	if err != nil {
 		return nil, fmt.Errorf("looking up %s: %w", repo, err)
 	}
 	d := inspectModel(ctx, m, cfg)
-	v, ok := catalog.FindVariant(d.All, want)
+	v, ok := resolve.FindVariant(d.All, want)
 	if !ok {
 		// A repo holding one copy has nothing to choose between, and listing
 		// its single placeholder label back ("it has: default") reads as a
@@ -242,12 +242,12 @@ func resolveQuant(ctx context.Context, cfg *config.Config, repo, want string) (*
 	if v.GGUF && !d.GGUFReady {
 		return nil, errors.New(ggufMissingMessage(cfg))
 	}
-	return &catalog.Choice{Variant: v, BaseModel: d.BaseModel}, nil
+	return &resolve.Choice{Variant: v, BaseModel: d.BaseModel}, nil
 }
 
 // pullAndSwitch is `new` minus the flag parsing, so `search` can funnel
 // into it with its own options.
-func pullAndSwitch(ctx context.Context, cfg *config.Config, repo string, choice *catalog.Choice, opts watchOpts) error {
+func pullAndSwitch(ctx context.Context, cfg *config.Config, repo string, choice *resolve.Choice, opts watchOpts) error {
 	name := profile.NameFromRepo(repo)
 	if choice != nil {
 		if _, ok := choice.Variant.Selector(); ok {
@@ -263,7 +263,7 @@ func pullAndSwitch(ctx context.Context, cfg *config.Config, repo string, choice
 	}
 
 	fmt.Println("Pulling new model:", repo)
-	prov, err := provider.Lookup(cfg.Provider())
+	prov, err := engine.Lookup(cfg.Provider())
 	if err != nil {
 		return err
 	}
@@ -274,7 +274,7 @@ func pullAndSwitch(ctx context.Context, cfg *config.Config, repo string, choice
 		fmt.Println("chat template for known crash patterns...")
 		caps = vllmCaps(cfg)
 	}
-	plan := catalog.PrepareFor(ctx, cfg.Dirs, prov, repo, choice, caps, cfg.KVContextTarget())
+	plan := resolve.PrepareFor(ctx, cfg.Dirs, prov, repo, choice, caps, cfg.KVContextTarget())
 	for _, n := range plan.Notes {
 		fmt.Println(n)
 	}
diff --git a/internal/cli/recipehealth.go b/internal/cli/recipehealth.go
index a7c8672..88b7042 100644
--- a/internal/cli/recipehealth.go
+++ b/internal/cli/recipehealth.go
@@ -11,8 +11,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // Two consecutive misses, two minutes apart, before relaunching. One miss is
diff --git a/internal/cli/recipehealth_internal_test.go b/internal/cli/recipehealth_internal_test.go
index 2c4cacf..94fc3e9 100644
--- a/internal/cli/recipehealth_internal_test.go
+++ b/internal/cli/recipehealth_internal_test.go
@@ -5,7 +5,7 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // The watchdog's whole job is deciding when a silent endpoint means trouble.
diff --git a/internal/cli/recipelaunch.go b/internal/cli/recipelaunch.go
index 7eaa3ad..345aed5 100644
--- a/internal/cli/recipelaunch.go
+++ b/internal/cli/recipelaunch.go
@@ -13,13 +13,13 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 // discoverScript reports the cluster interface, both RoCE devices and this
@@ -589,8 +589,8 @@ func newRecipeProfile(dirs config.Dirs, r recipe.TestedRecipe, limit string) (*p
 	}
 	// An engine that takes vLLM's args keeps the sizing keys, and is still
 	// named.
-	if e.ID != provider.VLLM {
-		p.File.Set(provider.Key, e.ID)
+	if e.ID != engine.VLLM {
+		p.File.Set(engine.Key, e.ID)
 	}
 	return p, nil
 }
diff --git a/internal/cli/recipelaunch_internal_test.go b/internal/cli/recipelaunch_internal_test.go
index f507115..9ee3086 100644
--- a/internal/cli/recipelaunch_internal_test.go
+++ b/internal/cli/recipelaunch_internal_test.go
@@ -8,10 +8,10 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 // Locality decides the hop. Before this was a property of the node, every
@@ -262,10 +262,10 @@ func TestNewRecipeProfileNamesTheEngine(t *testing.T) {
 	if err := os.WriteFile(filepath.Join(dir, "toyv.json"), []byte(toy), 0o644); err != nil {
 		t.Fatal(err)
 	}
-	if errs := provider.UseLocal(dir); len(errs) > 0 {
+	if errs := engine.UseLocal(dir); len(errs) > 0 {
 		t.Fatal(errs)
 	}
-	t.Cleanup(func() { provider.UseLocal(filepath.Join(dir, "gone")) })
+	t.Cleanup(func() { engine.UseLocal(filepath.Join(dir, "gone")) })
 
 	vllmLike := testRecipe(t)
 	vllmLike.ID = "toyv-recipe"
diff --git a/internal/cli/recipes.go b/internal/cli/recipes.go
index 2eeed3c..47d2360 100644
--- a/internal/cli/recipes.go
+++ b/internal/cli/recipes.go
@@ -12,7 +12,7 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 func newRecipesCmd() *cobra.Command {
diff --git a/internal/cli/recipes_internal_test.go b/internal/cli/recipes_internal_test.go
index e6221b6..82d3a4e 100644
--- a/internal/cli/recipes_internal_test.go
+++ b/internal/cli/recipes_internal_test.go
@@ -5,7 +5,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 // The display name column was a fixed 22 wide, so a longer name such as
diff --git a/internal/cli/root.go b/internal/cli/root.go
index dec775c..aa56692 100644
--- a/internal/cli/root.go
+++ b/internal/cli/root.go
@@ -12,14 +12,14 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
-	"github.com/coder/vllm-helper/internal/service"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/service"
 )
 
 // Version is stamped by goreleaser via ldflags.
@@ -76,7 +76,7 @@ func newRootCmd() *cobra.Command {
 		hardware.UseLocal(filepath.Join(config.ResolveDirs().Config, hardware.LocalDir))
 		// Engines load after hardware, because a local engine may name a
 		// vendor only a local hardware definition adds.
-		provider.UseLocal(filepath.Join(config.ResolveDirs().Config, provider.LocalDir))
+		engine.UseLocal(filepath.Join(config.ResolveDirs().Config, engine.LocalDir))
 		// Recipes load last, because a local recipe may run on a local
 		// engine.
 		recipe.UseLocal(filepath.Join(config.ResolveDirs().Config, recipe.LocalDir))
diff --git a/internal/cli/run_test.go b/internal/cli/run_test.go
index d42ea10..317a176 100644
--- a/internal/cli/run_test.go
+++ b/internal/cli/run_test.go
@@ -15,7 +15,7 @@ import (
 	"sync"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/cli"
+	"github.com/coder/llm-helper/internal/cli"
 )
 
 const sampleProfile = `# Status: TESTED AND WORKING
@@ -814,7 +814,7 @@ func TestBinaryExitCodes(t *testing.T) {
 		t.Skip("no go toolchain to build the binary with")
 	}
 	bin := filepath.Join(t.TempDir(), "vllm-helper")
-	build := exec.Command("go", "build", "-o", bin, "github.com/coder/vllm-helper")
+	build := exec.Command("go", "build", "-o", bin, "github.com/coder/llm-helper")
 	if out, err := build.CombinedOutput(); err != nil {
 		t.Fatalf("go build: %v\n%s", err, out)
 	}
diff --git a/internal/cli/search.go b/internal/cli/search.go
index 6804f3b..44f5978 100644
--- a/internal/cli/search.go
+++ b/internal/cli/search.go
@@ -9,12 +9,12 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/catalog"
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/resolve"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // cmdSearch searches HuggingFace and walks the user through picking a
@@ -47,9 +47,9 @@ func runSearch(ctx context.Context, args []string, jsonOut bool) error {
 		return fmt.Errorf("search request failed (network or rate limit): %w", err)
 	}
 	if jsonOut {
-		out := make([]catalog.Result, 0, len(models))
+		out := make([]resolve.Result, 0, len(models))
 		for _, m := range models {
-			out = append(out, catalog.ToResult(m))
+			out = append(out, resolve.ToResult(m))
 		}
 		return printJSON(out)
 	}
@@ -152,14 +152,14 @@ func runInspect(ctx context.Context, args []string, jsonOut bool) error {
 // inspectModel measures a model against this machine, for the provider new
 // profiles are served by. cfg may be nil, which just means there is no budget
 // to measure against.
-func inspectModel(ctx context.Context, m hf.Model, cfg *config.Config) catalog.Detail {
+func inspectModel(ctx context.Context, m hf.Model, cfg *config.Config) resolve.Detail {
 	var override string
 	var target int
-	prov, _ := provider.Lookup("")
+	prov, _ := engine.Lookup("")
 	if cfg != nil {
 		override = cfg.ModelBudgetGB()
 		target = cfg.KVContextTarget()
-		if p, err := provider.Lookup(cfg.Provider()); err == nil {
+		if p, err := engine.Lookup(cfg.Provider()); err == nil {
 			prov = p
 		}
 	}
@@ -168,7 +168,7 @@ func inspectModel(ctx context.Context, m hf.Model, cfg *config.Config) catalog.D
 	if prov.Capabilities().VLLMArgs {
 		caps = vllmCaps(cfg)
 	}
-	return catalog.ForProvider(catalog.Inspect(ctx, m, detect.ModelBudget(override), caps, target), prov)
+	return resolve.ForProvider(resolve.Inspect(ctx, m, detect.ModelBudget(override), caps, target), prov)
 }
 
 // firstClause keeps the head of a reason. Repeating a full sentence on every
@@ -200,7 +200,7 @@ func thousands(n int64) string {
 	return fmt.Sprintf("%.0fk", float64(n)/1000)
 }
 
-func printModelDetail(d catalog.Detail, cfg *config.Config) {
+func printModelDetail(d resolve.Detail, cfg *config.Config) {
 	fmt.Println(d.ID)
 	fmt.Println(d.URL)
 	fmt.Printf("  license:    %s\n", d.License)
@@ -290,7 +290,7 @@ func printModelDetail(d catalog.Detail, cfg *config.Config) {
 // while its Q8_0 is 34.8 GB and fits an 84 GB budget nearly twice over. A
 // reader who only sees the smallest will write the repo off, or run a 2-bit
 // quant without ever knowing there was a choice.
-func printVariants(d catalog.Detail) {
+func printVariants(d resolve.Detail) {
 	if len(d.All) < 2 {
 		return
 	}
@@ -326,7 +326,7 @@ func printVariants(d catalog.Detail) {
 // The default is pre-selected and one keypress away, so the common case costs
 // nothing -- but a repo holding a 2-bit and an 8-bit copy should not decide
 // between them on the user's behalf and say nothing.
-func pickQuant(cfg *config.Config, d catalog.Detail) (choice *catalog.Choice, cancelled bool) {
+func pickQuant(cfg *config.Config, d resolve.Detail) (choice *resolve.Choice, cancelled bool) {
 	servable := make([]hf.Variant, 0, len(d.All))
 	for _, v := range d.All {
 		if v.Servable() && (d.GGUFReady || !v.GGUF) {
@@ -382,6 +382,6 @@ func pickQuant(cfg *config.Config, d catalog.Detail) (choice *catalog.Choice, ca
 		if _, ok := v.Selector(); !ok {
 			return nil, false
 		}
-		return &catalog.Choice{Variant: v, BaseModel: d.BaseModel}, false
+		return &resolve.Choice{Variant: v, BaseModel: d.BaseModel}, false
 	}
 }
diff --git a/internal/cli/serve.go b/internal/cli/serve.go
index 9ca1a8b..03dde50 100644
--- a/internal/cli/serve.go
+++ b/internal/cli/serve.go
@@ -12,7 +12,7 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/web"
+	"github.com/coder/llm-helper/internal/web"
 )
 
 func newServeCmd() *cobra.Command {
diff --git a/internal/cli/service.go b/internal/cli/service.go
index 96322db..ace27f8 100644
--- a/internal/cli/service.go
+++ b/internal/cli/service.go
@@ -6,8 +6,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func newStopCmd() *cobra.Command {
diff --git a/internal/cli/setup.go b/internal/cli/setup.go
index d53a555..04ee20e 100644
--- a/internal/cli/setup.go
+++ b/internal/cli/setup.go
@@ -13,9 +13,9 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
 )
 
 func newSetupCmd() *cobra.Command {
@@ -27,8 +27,8 @@ func newSetupCmd() *cobra.Command {
 		Args:    cobra.NoArgs,
 		RunE:    func(*cobra.Command, []string) error { return runSetup(providerName) },
 	}
-	c.Flags().StringVar(&providerName, "provider", provider.VLLM,
-		"inference server new profiles use: "+strings.Join(provider.ProcessNames(), ", ")+" (only written to a new config)")
+	c.Flags().StringVar(&providerName, "provider", engine.VLLM,
+		"inference server new profiles use: "+strings.Join(engine.ProcessNames(), ", ")+" (only written to a new config)")
 	return c
 }
 
@@ -37,11 +37,11 @@ func newSetupCmd() *cobra.Command {
 // offer to adopt a running server. Idempotent; never overwrites existing
 // config and never restarts anything.
 func runSetup(providerName string) error {
-	prov, err := provider.Lookup(providerName)
+	prov, err := engine.Lookup(providerName)
 	if err != nil {
 		return err
 	}
-	if !prov.Has(provider.RuntimeProcess) {
+	if !prov.Has(engine.RuntimeProcess) {
 		return fmt.Errorf("%s runs only in a container, so the service cannot serve it; launch it with 'llm-helper recipes launch'", prov.Name())
 	}
 	dirs := config.ResolveDirs()
@@ -54,7 +54,7 @@ func runSetup(providerName string) error {
 	bin := ""
 	if statErr != nil {
 		bin = lookBin(prov)
-		if prov.Name() == provider.VLLM {
+		if prov.Name() == engine.VLLM {
 			useMetalProbe(bin)
 		}
 	} else if found, filled, err := fillMissingVLLMBin(dirs.ConfigFile(), findVLLM); err != nil {
@@ -92,7 +92,7 @@ func runSetup(providerName string) error {
 
 	if statErr == nil {
 		fmt.Println("Config already exists at", dirs.ConfigFile(), "-- keeping it.")
-		if prov.Name() != provider.VLLM {
+		if prov.Name() != engine.VLLM {
 			fmt.Printf("--provider only shapes a new config; set PROVIDER=%s and %s in it by hand.\n", prov.Name(), prov.BinKey())
 		}
 	} else {
@@ -133,7 +133,7 @@ TIKTOKEN_ENCODINGS_BASE=%s
 #CODER_AI_PROVIDER_ID=
 # Only needed if the deployment has more than one organization.
 #CODER_ORGANIZATION=default
-`, strings.Join(provider.ProcessNames(), ", "), prov.Name(), prov.BinName(), prov.BinKey(), bin,
+`, strings.Join(engine.ProcessNames(), ", "), prov.Name(), prov.BinName(), prov.BinKey(), bin,
 			config.ServiceManager, config.DefaultServiceScope, config.DefaultServiceName,
 			hw.BudgetGB, hw.BudgetReason, hw.BudgetGB, filepath.Join(dirs.Data, "tiktoken-encodings"),
 			detect.DefaultContextTarget, detect.DefaultContextTarget)
@@ -143,7 +143,7 @@ TIKTOKEN_ENCODINGS_BASE=%s
 		fmt.Println("Wrote", dirs.ConfigFile())
 		if bin == "" {
 			fmt.Println()
-			if prov.Name() == provider.VLLM {
+			if prov.Name() == engine.VLLM {
 				for _, line := range vllmInstallHint {
 					fmt.Println(line)
 				}
@@ -152,8 +152,8 @@ TIKTOKEN_ENCODINGS_BASE=%s
 				fmt.Printf("'%s'), then set %s\n", prov.InstallHint(runtime.GOOS), prov.BinKey())
 				fmt.Println("in config.env to its full path.")
 			}
-			for _, name := range provider.ProcessNames() {
-				other, _ := provider.Lookup(name)
+			for _, name := range engine.ProcessNames() {
+				other, _ := engine.Lookup(name)
 				if name == prov.Name() || lookBin(other) == "" {
 					continue
 				}
@@ -234,7 +234,7 @@ func fillMissingVLLMBin(path string, find func() string) (string, bool, error) {
 	if err != nil {
 		return "", false, fmt.Errorf("reading %s: %w", path, err)
 	}
-	if prov, err := provider.Lookup(f.Get(provider.Key)); err != nil || prov.Name() != provider.VLLM {
+	if prov, err := engine.Lookup(f.Get(engine.Key)); err != nil || prov.Name() != engine.VLLM {
 		return "", false, nil
 	}
 	if strings.TrimSpace(f.Get("VLLM_BIN")) != "" {
@@ -256,8 +256,8 @@ func fillMissingVLLMBin(path string, find func() string) (string, bool, error) {
 // up beside it. Any other server is kept as found on PATH, so a package
 // manager's stable link is what config.env records and an upgrade does not
 // break it.
-func lookBin(prov provider.Provider) string {
-	if prov.Name() == provider.VLLM {
+func lookBin(prov engine.Provider) string {
+	if prov.Name() == engine.VLLM {
 		return findVLLM()
 	}
 	p, err := exec.LookPath(prov.BinName())
diff --git a/internal/cli/shard_internal_test.go b/internal/cli/shard_internal_test.go
index 6688f54..2ae9fc8 100644
--- a/internal/cli/shard_internal_test.go
+++ b/internal/cli/shard_internal_test.go
@@ -8,8 +8,8 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/nodes"
 )
 
 func TestShardRayEnvPrefixesUseNodeLocalHostIP(t *testing.T) {
@@ -108,7 +108,7 @@ func TestShardMPServeArgs(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	spec := serveSpec{Spec: provider.Spec{
+	spec := serveSpec{Spec: engine.Spec{
 		Bin:            bin,
 		Model:          "org/model",
 		Port:           "8021",
diff --git a/internal/cli/shard_mp.go b/internal/cli/shard_mp.go
index a6e2242..632efd4 100644
--- a/internal/cli/shard_mp.go
+++ b/internal/cli/shard_mp.go
@@ -11,9 +11,9 @@ import (
 	"strings"
 	"syscall"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func normalizeShardBackend(backend string, allowAll bool) (string, error) {
diff --git a/internal/cli/shardcmd.go b/internal/cli/shardcmd.go
index b424777..87fac37 100644
--- a/internal/cli/shardcmd.go
+++ b/internal/cli/shardcmd.go
@@ -11,9 +11,9 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func newShardCmd() *cobra.Command {
diff --git a/internal/cli/start.go b/internal/cli/start.go
index d4ab096..d364095 100644
--- a/internal/cli/start.go
+++ b/internal/cli/start.go
@@ -9,8 +9,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func newStartCmd() *cobra.Command {
diff --git a/internal/cli/upgrade.go b/internal/cli/upgrade.go
index 963bf0a..738f4bf 100644
--- a/internal/cli/upgrade.go
+++ b/internal/cli/upgrade.go
@@ -9,8 +9,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/update"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/update"
 )
 
 // updateChannelKey selects where upgrades come from. It lives in config.env
diff --git a/internal/cli/upgrade_internal_test.go b/internal/cli/upgrade_internal_test.go
index ecf4cbb..d5acc03 100644
--- a/internal/cli/upgrade_internal_test.go
+++ b/internal/cli/upgrade_internal_test.go
@@ -5,8 +5,8 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/update"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/update"
 )
 
 // Upgrading has to work on a machine that has never run setup -- that is
diff --git a/internal/cli/vllmargs.go b/internal/cli/vllmargs.go
index d4b5815..9b37363 100644
--- a/internal/cli/vllmargs.go
+++ b/internal/cli/vllmargs.go
@@ -5,27 +5,27 @@ import (
 	"os"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/recipe"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/recipe"
 )
 
 type serveSpec struct {
 	// Provider is nil for vLLM, so a spec built without one keeps meaning
 	// what it meant before there was a choice.
-	Provider provider.Provider
-	provider.Spec
+	Provider engine.Provider
+	engine.Spec
 }
 
 func serveSpecFromActive(active *config.EnvFile) (serveSpec, error) {
-	prov, err := provider.Lookup(active.Get(provider.Key))
+	prov, err := engine.Lookup(active.Get(engine.Key))
 	if err != nil {
 		return serveSpec{}, err
 	}
-	return serveSpec{Provider: prov, Spec: provider.Spec{
+	return serveSpec{Provider: prov, Spec: engine.Spec{
 		Bin:            active.Get(prov.BinKey()),
 		Model:          active.Get("MODEL"),
 		Port:           active.Get("PORT"),
@@ -45,7 +45,7 @@ func serveSpecFromProfile(cfg *config.Config, p *profile.Profile, port string) (
 	if err != nil {
 		return serveSpec{}, err
 	}
-	return serveSpec{Provider: prov, Spec: provider.Spec{
+	return serveSpec{Provider: prov, Spec: engine.Spec{
 		Bin:            cfg.Lookup(prov.BinKey(), ""),
 		Model:          p.Model(),
 		Port:           port,
@@ -61,20 +61,20 @@ func serveSpecFromProfile(cfg *config.Config, p *profile.Profile, port string) (
 // own PROVIDER counts: config.env's names the default for new profiles, and
 // letting it reach an existing one would hand a vLLM profile's flags to
 // another server.
-func profileProvider(p *profile.Profile) (provider.Provider, error) {
-	return provider.Lookup(p.Provider())
+func profileProvider(p *profile.Profile) (engine.Provider, error) {
+	return engine.Lookup(p.Provider())
 }
 
-func (s serveSpec) provider() provider.Provider {
+func (s serveSpec) provider() engine.Provider {
 	if s.Provider != nil {
 		return s.Provider
 	}
-	p, _ := provider.Lookup(provider.VLLM)
+	p, _ := engine.Lookup(engine.VLLM)
 	return p
 }
 
 func (s serveSpec) validate() error {
-	return provider.Validate(s.provider(), s.Spec)
+	return engine.Validate(s.provider(), s.Spec)
 }
 
 // args is the provider's command line with extra appended. Validation comes
@@ -177,7 +177,7 @@ func withPCIBusOrder(env []string) []string {
 // replaces an inherited entry of the same name, because a duplicate would
 // leave the server reading whichever copy its libc finds first.
 func withProviderEnv(env []string, s serveSpec) []string {
-	ep, ok := s.provider().(provider.EnvProvider)
+	ep, ok := s.provider().(engine.EnvProvider)
 	if !ok {
 		return env
 	}
diff --git a/internal/cli/vllmargs_internal_test.go b/internal/cli/vllmargs_internal_test.go
index 78075f6..aa7c4b0 100644
--- a/internal/cli/vllmargs_internal_test.go
+++ b/internal/cli/vllmargs_internal_test.go
@@ -7,10 +7,10 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func TestRedactServeArgs(t *testing.T) {
@@ -153,17 +153,17 @@ func TestRequireVLLM(t *testing.T) {
 // inherited, and a vLLM spec's environment is left as it was.
 func TestWithProviderEnv(t *testing.T) {
 	t.Parallel()
-	llama, err := provider.Lookup(provider.LlamaCPP)
+	llama, err := engine.Lookup(engine.LlamaCPP)
 	if err != nil {
 		t.Fatal(err)
 	}
 	env := []string{"PATH=/usr/bin", "LLAMA_API_KEY=stale", "VLLM_API_KEY=sk-x"}
-	got := withProviderEnv(env, serveSpec{Provider: llama, Spec: provider.Spec{APIKey: "sk-x"}})
+	got := withProviderEnv(env, serveSpec{Provider: llama, Spec: engine.Spec{APIKey: "sk-x"}})
 	want := []string{"PATH=/usr/bin", "VLLM_API_KEY=sk-x", "LLAMA_API_KEY=sk-x"}
 	if !reflect.DeepEqual(got, want) {
 		t.Errorf("llamacpp env = %q, want %q", got, want)
 	}
-	if got := withProviderEnv(env, serveSpec{Spec: provider.Spec{APIKey: "sk-x"}}); !reflect.DeepEqual(got, env) {
+	if got := withProviderEnv(env, serveSpec{Spec: engine.Spec{APIKey: "sk-x"}}); !reflect.DeepEqual(got, env) {
 		t.Errorf("vllm env = %q, want it unchanged", got)
 	}
 }
@@ -184,7 +184,7 @@ func TestServeArgsRefuseAFlagThisArchitectureAbortsOn(t *testing.T) {
 	if err := os.WriteFile(bin, []byte("#!/bin/sh\n"), 0o755); err != nil {
 		t.Fatal(err)
 	}
-	spec := serveSpec{Spec: provider.Spec{
+	spec := serveSpec{Spec: engine.Spec{
 		Bin: bin, Model: "m", Port: "8001",
 		ExtraArgs: "--moe-backend deep_gemm_mega_moe",
 	}}
@@ -208,7 +208,7 @@ func TestServeArgsRefuseAFlagThisArchitectureAbortsOn(t *testing.T) {
 // A shard's workers run the head's command line on their own hardware, so a
 // worker the registry knows is a GB10 is checked before any node is touched.
 func TestCheckNodesRefusesAWorkerThatWouldAbort(t *testing.T) {
-	spec := serveSpec{Spec: provider.Spec{
+	spec := serveSpec{Spec: engine.Spec{
 		Bin: "/opt/vllm/bin/vllm", Model: "m", Port: "8001",
 		ExtraArgs: "--moe-backend deep_gemm_mega_moe",
 	}}
@@ -229,7 +229,7 @@ func TestCheckNodesRefusesAWorkerThatWouldAbort(t *testing.T) {
 // architecture for it, or a stale one. Checking it only when the command was
 // built let --stop-services stop every node's server first.
 func TestCheckNodesChecksTheHeadAgainstThisMachine(t *testing.T) {
-	spec := serveSpec{Spec: provider.Spec{
+	spec := serveSpec{Spec: engine.Spec{
 		Bin: "/opt/vllm/bin/vllm", Model: "m", Port: "8001",
 		ExtraArgs: "--moe-backend deep_gemm_mega_moe",
 	}}
diff --git a/internal/cli/vllmbin_internal_test.go b/internal/cli/vllmbin_internal_test.go
index 0f3efb5..505af85 100644
--- a/internal/cli/vllmbin_internal_test.go
+++ b/internal/cli/vllmbin_internal_test.go
@@ -6,7 +6,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // /opt/homebrew/bin/vllm resolves into a versioned Cellar directory that the
diff --git a/internal/cli/watch.go b/internal/cli/watch.go
index 52ef4d1..529c598 100644
--- a/internal/cli/watch.go
+++ b/internal/cli/watch.go
@@ -6,8 +6,8 @@ import (
 	"os"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
 )
 
 // watchOpts controls how a restart is observed.
diff --git a/internal/cli/watch_internal_test.go b/internal/cli/watch_internal_test.go
index 102d150..c646c51 100644
--- a/internal/cli/watch_internal_test.go
+++ b/internal/cli/watch_internal_test.go
@@ -12,9 +12,9 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/nodes"
-	"github.com/coder/vllm-helper/internal/service"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/nodes"
+	"github.com/coder/llm-helper/internal/service"
 )
 
 // The boot wait has to ask the machine that serves. A cluster's head can be
diff --git a/internal/config/config_test.go b/internal/config/config_test.go
index 0e23aa3..e9d1c60 100644
--- a/internal/config/config_test.go
+++ b/internal/config/config_test.go
@@ -6,7 +6,7 @@ import (
 	"reflect"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // writeFile creates a file and the directories leading to it.
diff --git a/internal/config/shellwords_test.go b/internal/config/shellwords_test.go
index df7f691..9762a74 100644
--- a/internal/config/shellwords_test.go
+++ b/internal/config/shellwords_test.go
@@ -4,7 +4,7 @@ import (
 	"reflect"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 func TestSplitWords(t *testing.T) {
diff --git a/internal/control/control.go b/internal/control/control.go
index 20c1b27..404ee41 100644
--- a/internal/control/control.go
+++ b/internal/control/control.go
@@ -19,10 +19,10 @@ import (
 	"strings"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // BaseURL is where the model server answers on this machine.
@@ -46,7 +46,7 @@ func BaseURL(cfg *config.Config) string {
 func WriteActiveEnv(cfg *config.Config, p *profile.Profile) error {
 	// An unknown provider would otherwise surface only when the service tried
 	// to start it, by which time the outgoing model is already down.
-	if _, err := provider.Lookup(p.Provider()); err != nil {
+	if _, err := engine.Lookup(p.Provider()); err != nil {
 		return fmt.Errorf("profile '%s': %w", p.Name, err)
 	}
 	if err := os.MkdirAll(cfg.Dirs.State, 0o755); err != nil {
@@ -75,7 +75,7 @@ func WriteActiveEnv(cfg *config.Config, p *profile.Profile) error {
 	// default to override, so an install that never set one gets the file
 	// it always got.
 	if p.Provider() == "" && cfg.Provider() != "" {
-		b.WriteString(provider.Key + "=" + provider.VLLM + "\n")
+		b.WriteString(engine.Key + "=" + engine.VLLM + "\n")
 	}
 
 	// Written and renamed rather than truncated in place: a crash halfway
@@ -182,7 +182,7 @@ func ActiveHealthPath(cfg *config.Config) string {
 	if err != nil {
 		return ""
 	}
-	e, err := provider.LookupEngine(active.Get(provider.Key))
+	e, err := engine.LookupEngine(active.Get(engine.Key))
 	if err != nil {
 		return ""
 	}
diff --git a/internal/control/control_internal_test.go b/internal/control/control_internal_test.go
index 8b5e461..1ee9c34 100644
--- a/internal/control/control_internal_test.go
+++ b/internal/control/control_internal_test.go
@@ -6,9 +6,9 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // active.env is sourced by the service, so what goes into it decides the argv
diff --git a/internal/control/gguf.go b/internal/control/gguf.go
index ddc5f81..ef522b7 100644
--- a/internal/control/gguf.go
+++ b/internal/control/gguf.go
@@ -9,8 +9,8 @@ import (
 	"os/exec"
 	"path/filepath"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // GGUF support left vLLM core in 0.27.1 and lives in an out-of-tree package.
diff --git a/internal/control/gpus.go b/internal/control/gpus.go
index 8523819..267f632 100644
--- a/internal/control/gpus.go
+++ b/internal/control/gpus.go
@@ -3,8 +3,8 @@ package control
 import (
 	"os"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // CUDAVisible is the CUDA_VISIBLE_DEVICES a server started with these env
diff --git a/internal/detect/classify.go b/internal/detect/classify.go
index d60bb02..04fbfd1 100644
--- a/internal/detect/classify.go
+++ b/internal/detect/classify.go
@@ -5,7 +5,7 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // What decided whether a GPU's memory is shared with the CPU, strongest
diff --git a/internal/detect/container_test.go b/internal/detect/container_test.go
index 6d82784..c0c7634 100644
--- a/internal/detect/container_test.go
+++ b/internal/detect/container_test.go
@@ -5,7 +5,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 func TestScanContainersAgainstTheRealRuntime(t *testing.T) {
diff --git a/internal/detect/endpoint_test.go b/internal/detect/endpoint_test.go
index e7ae715..f0dd1af 100644
--- a/internal/detect/endpoint_test.go
+++ b/internal/detect/endpoint_test.go
@@ -8,7 +8,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // serveOn starts an OpenAI-shaped server and returns the port it listens on.
diff --git a/internal/detect/fixtures_internal_test.go b/internal/detect/fixtures_internal_test.go
index 97e54cd..15a8216 100644
--- a/internal/detect/fixtures_internal_test.go
+++ b/internal/detect/fixtures_internal_test.go
@@ -9,7 +9,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // fixtureDir holds the hardware reports the definitions are tested against.
diff --git a/internal/detect/hardware_internal_test.go b/internal/detect/hardware_internal_test.go
index 55e9244..441d325 100644
--- a/internal/detect/hardware_internal_test.go
+++ b/internal/detect/hardware_internal_test.go
@@ -4,7 +4,7 @@ import (
 	"testing"
 	"testing/fstest"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // unifiedFor classifies one NVIDIA GPU the way the probe does.
diff --git a/internal/detect/hardware_test.go b/internal/detect/hardware_test.go
index d6d3f89..ffc2e1f 100644
--- a/internal/detect/hardware_test.go
+++ b/internal/detect/hardware_test.go
@@ -5,7 +5,7 @@ package detect_test
 import (
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 // The budget drives whether `search` calls a model too big, so a zero here
diff --git a/internal/detect/process.go b/internal/detect/process.go
index 232919e..892257a 100644
--- a/internal/detect/process.go
+++ b/internal/detect/process.go
@@ -6,7 +6,7 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // Server is one detected vLLM server and everything we learned about it.
diff --git a/internal/detect/process_internal_test.go b/internal/detect/process_internal_test.go
index 47502a3..f1680ba 100644
--- a/internal/detect/process_internal_test.go
+++ b/internal/detect/process_internal_test.go
@@ -5,7 +5,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // Real-world argv observed on a GB10 box (venv console-script form).
diff --git a/internal/detect/report.go b/internal/detect/report.go
index 949bd81..0cac2b7 100644
--- a/internal/detect/report.go
+++ b/internal/detect/report.go
@@ -1,7 +1,7 @@
 package detect
 
 import (
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // Report is what 'hardware report' prints and what a hardware fixture holds:
diff --git a/internal/provider/engine.go b/internal/engine/engine.go
similarity index 99%
rename from internal/provider/engine.go
rename to internal/engine/engine.go
index c9ab999..2030457 100644
--- a/internal/provider/engine.go
+++ b/internal/engine/engine.go
@@ -1,4 +1,4 @@
-package provider
+package engine
 
 import (
 	"bytes"
@@ -13,9 +13,9 @@ import (
 	"slices"
 	"strings"
 
-	"github.com/coder/vllm-helper/catalogue"
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/catalogue"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // SchemaVersion is the engine definitions format this build reads. A file
diff --git a/internal/provider/engine_test.go b/internal/engine/engine_test.go
similarity index 80%
rename from internal/provider/engine_test.go
rename to internal/engine/engine_test.go
index 649bd00..6b3ca70 100644
--- a/internal/provider/engine_test.go
+++ b/internal/engine/engine_test.go
@@ -1,4 +1,4 @@
-package provider_test
+package engine_test
 
 import (
 	"io/fs"
@@ -9,8 +9,8 @@ import (
 	"testing"
 	"testing/fstest"
 
-	"github.com/coder/vllm-helper/catalogue"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/catalogue"
+	"github.com/coder/llm-helper/internal/engine"
 )
 
 const secretKey = "sk-never-on-argv"
@@ -20,15 +20,15 @@ const secretKey = "sk-never-on-argv"
 // user on the machine.
 func TestShippedEnginesHoldTheContract(t *testing.T) {
 	t.Parallel()
-	for _, e := range provider.Shipped().Engines() {
+	for _, e := range engine.Shipped().Engines() {
 		if strings.TrimSpace(e.Why) == "" {
 			t.Errorf("%s: no why", e.ID)
 		}
-		if e.Has(provider.RuntimeProcess) {
+		if e.Has(engine.RuntimeProcess) {
 			if e.InstallHint("linux") == "" || e.InstallHint("darwin") == "" {
 				t.Errorf("%s: a process engine needs an install hint on every OS", e.ID)
 			}
-			argv, err := e.Command(provider.Spec{Bin: "/x", Model: "org/m", Port: "8001", APIKey: secretKey})
+			argv, err := e.Command(engine.Spec{Bin: "/x", Model: "org/m", Port: "8001", APIKey: secretKey})
 			if err != nil {
 				t.Fatalf("%s: %v", e.ID, err)
 			}
@@ -42,7 +42,7 @@ func TestShippedEnginesHoldTheContract(t *testing.T) {
 				}
 			}
 		}
-		if !e.Has(provider.RuntimeContainer) {
+		if !e.Has(engine.RuntimeContainer) {
 			continue
 		}
 		serving := 0
@@ -54,16 +54,16 @@ func TestShippedEnginesHoldTheContract(t *testing.T) {
 		if serving != 1 {
 			t.Errorf("%s: %d roles serve HTTP, want 1", e.ID, serving)
 		}
-		models := map[string]provider.ModelRef{}
+		models := map[string]engine.ModelRef{}
 		for _, role := range e.UsesModelPaths() {
-			models[role] = provider.ModelRef{ID: "org/" + role, Revision: strings.Repeat("a", 40)}
+			models[role] = engine.ModelRef{ID: "org/" + role, Revision: strings.Repeat("a", 40)}
 		}
 		for rank := range 4 {
 			r, err := e.RoleFor(rank)
 			if err != nil {
 				t.Fatalf("%s: %v", e.ID, err)
 			}
-			_, args, env := e.RenderRole(r, provider.RoleValues{
+			_, args, env := e.RenderRole(r, engine.RoleValues{
 				Model: "m", Port: 8000, Rank: rank, Nodes: 4, MasterAddr: "10.0.0.1", MasterPort: 29500,
 				Args: []string{"--x", "1"}, Models: models,
 			})
@@ -91,11 +91,11 @@ func TestShippedEnginesHoldTheContract(t *testing.T) {
 // CUDA, so neither may be overridden.
 func TestTensorFoldPythonKeepsTheImageEntrypointAndTakesNoKey(t *testing.T) {
 	t.Parallel()
-	e, ok := provider.Shipped().Get("tensorfold-python")
+	e, ok := engine.Shipped().Get("tensorfold-python")
 	if !ok {
 		t.Fatal("tensorfold-python is not shipped")
 	}
-	if e.Has(provider.RuntimeProcess) {
+	if e.Has(engine.RuntimeProcess) {
 		t.Error("tensorfold-python runs only in a container")
 	}
 	if e.Capabilities().APIKey || e.ContainerKeyEnv() != "" {
@@ -103,7 +103,7 @@ func TestTensorFoldPythonKeepsTheImageEntrypointAndTakesNoKey(t *testing.T) {
 	}
 	for rank := range 2 {
 		r, _ := e.RoleFor(rank)
-		if ep, _, _ := e.RenderRole(r, provider.RoleValues{}); ep != "" {
+		if ep, _, _ := e.RenderRole(r, engine.RoleValues{}); ep != "" {
 			t.Errorf("rank %d overrides the image entrypoint with %q", rank, ep)
 		}
 	}
@@ -154,7 +154,7 @@ const toyProcess = `{
 func TestParseIsStrict(t *testing.T) {
 	t.Parallel()
 	for _, base := range []string{toyContainer, toyProcess} {
-		if _, err := provider.Parse([]byte(base)); err != nil {
+		if _, err := engine.Parse([]byte(base)); err != nil {
 			t.Fatalf("the base definition does not parse: %v", err)
 		}
 	}
@@ -182,7 +182,7 @@ func TestParseIsStrict(t *testing.T) {
 		if data == tc.base && tc.name != "trailing data" {
 			t.Fatalf("%s: the edit matched nothing", tc.name)
 		}
-		_, err := provider.Parse([]byte(data))
+		_, err := engine.Parse([]byte(data))
 		if err == nil || !strings.Contains(err.Error(), tc.want) {
 			t.Errorf("%s: err = %v, want it to mention %q", tc.name, err, tc.want)
 		}
@@ -194,12 +194,12 @@ func TestParseIsStrict(t *testing.T) {
 func TestLaunchArgsFillSharedValues(t *testing.T) {
 	t.Parallel()
 	data := strings.Replace(toyContainer, `"api_key": "none"`, `"api_key": "none", "launch_args": ["--served-model-name", "{model}", "--threads", "{threads}"]`, 1)
-	r, err := provider.Parse([]byte(data))
+	r, err := engine.Parse([]byte(data))
 	if err != nil {
 		t.Fatal(err)
 	}
 	e, _ := r.Get("toy")
-	got := e.LaunchArgs(provider.RoleValues{Model: "org/m", Port: 8000})
+	got := e.LaunchArgs(engine.RoleValues{Model: "org/m", Port: 8000})
 	if want := []string{"--served-model-name", "org/m", "--threads", "4"}; !slices.Equal(got, want) {
 		t.Errorf("LaunchArgs = %v, want %v", got, want)
 	}
@@ -209,7 +209,7 @@ func TestLaunchArgsFillSharedValues(t *testing.T) {
 // chat template's {{bos_token}} is not a placeholder.
 func TestRenderRoleLeavesArgsAlone(t *testing.T) {
 	t.Parallel()
-	r, err := provider.Parse([]byte(toyContainer))
+	r, err := engine.Parse([]byte(toyContainer))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -219,7 +219,7 @@ func TestRenderRoleLeavesArgsAlone(t *testing.T) {
 		t.Fatal(err)
 	}
 	tmpl := "{{bos_token}}{model}"
-	_, args, _ := e.RenderRole(role, provider.RoleValues{Model: "org/m", Port: 8000, Nodes: 1, Args: []string{"--chat-template", tmpl}})
+	_, args, _ := e.RenderRole(role, engine.RoleValues{Model: "org/m", Port: 8000, Nodes: 1, Args: []string{"--chat-template", tmpl}})
 	if got := args[len(args)-1]; got != tmpl {
 		t.Errorf("--chat-template = %q, want %q as written", got, tmpl)
 	}
@@ -227,7 +227,7 @@ func TestRenderRoleLeavesArgsAlone(t *testing.T) {
 
 func TestExtendOnlyAdds(t *testing.T) {
 	t.Parallel()
-	r, err := provider.Shipped().Extend([]byte(toyContainer), "/etc/toy.json")
+	r, err := engine.Shipped().Extend([]byte(toyContainer), "/etc/toy.json")
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -235,19 +235,19 @@ func TestExtendOnlyAdds(t *testing.T) {
 	if !ok || e.Source != "/etc/toy.json" {
 		t.Fatalf("toy = %+v, %v", e, ok)
 	}
-	if _, ok := provider.Shipped().Get("toy"); ok {
+	if _, ok := engine.Shipped().Get("toy"); ok {
 		t.Error("Extend changed the registry it extended")
 	}
 
 	redefine := strings.Replace(toyContainer, `"id": "toy"`, `"id": "vllm"`, 1)
-	if _, err := provider.Shipped().Extend([]byte(redefine), "x.json"); err == nil || !strings.Contains(err.Error(), "already defined") {
+	if _, err := engine.Shipped().Extend([]byte(redefine), "x.json"); err == nil || !strings.Contains(err.Error(), "already defined") {
 		t.Errorf("redefining vllm: err = %v", err)
 	}
 	invalid := strings.Replace(toyContainer, `"{threads}"`, `"{nope}"`, 1)
-	if _, err := provider.Shipped().Extend([]byte(invalid), "toy.json"); err == nil {
+	if _, err := engine.Shipped().Extend([]byte(invalid), "toy.json"); err == nil {
 		t.Error("an invalid local engine was accepted")
 	}
-	if _, err := provider.Shipped().Extend([]byte(toyContainer), "/etc/other.json"); err == nil || !strings.Contains(err.Error(), "toy.json") {
+	if _, err := engine.Shipped().Extend([]byte(toyContainer), "/etc/other.json"); err == nil || !strings.Contains(err.Error(), "toy.json") {
 		t.Errorf("a file named for another engine: err = %v", err)
 	}
 }
@@ -255,10 +255,10 @@ func TestExtendOnlyAdds(t *testing.T) {
 // Load refuses a folder whose file names and ids disagree, naming the file.
 func TestLoadHoldsFilesToTheirIDs(t *testing.T) {
 	t.Parallel()
-	if _, err := provider.Load(fstest.MapFS{"toy.json": {Data: []byte(toyContainer)}}); err != nil {
+	if _, err := engine.Load(fstest.MapFS{"toy.json": {Data: []byte(toyContainer)}}); err != nil {
 		t.Fatalf("a well-named folder: %v", err)
 	}
-	_, err := provider.Load(fstest.MapFS{"toy.json": {Data: []byte(toyContainer)}, "box.json": {Data: []byte(toyProcess)}})
+	_, err := engine.Load(fstest.MapFS{"toy.json": {Data: []byte(toyContainer)}, "box.json": {Data: []byte(toyProcess)}})
 	if err == nil || !strings.Contains(err.Error(), "box.json") || !strings.Contains(err.Error(), "toyp.json") {
 		t.Errorf("a misnamed file: err = %v", err)
 	}
@@ -273,7 +273,7 @@ func TestShippedEnginesAreOneFileEach(t *testing.T) {
 		t.Fatal(err)
 	}
 	var ids []string
-	for _, e := range provider.Shipped().Engines() {
+	for _, e := range engine.Shipped().Engines() {
 		ids = append(ids, e.ID+".json")
 	}
 	if !slices.Equal(names, ids) {
@@ -297,9 +297,9 @@ func TestUseLocalSkipsAndReportsBadFiles(t *testing.T) {
 			t.Fatal(err)
 		}
 	}
-	t.Cleanup(func() { provider.UseLocal(filepath.Join(t.TempDir(), "gone")) })
+	t.Cleanup(func() { engine.UseLocal(filepath.Join(t.TempDir(), "gone")) })
 
-	errs := provider.UseLocal(dir)
+	errs := engine.UseLocal(dir)
 	// A file is named after the engine it defines, so the misnamed one is
 	// refused too.
 	wantErrs := []string{"broken.json", "misnamed.json", "vllm.json"}
@@ -311,29 +311,29 @@ func TestUseLocalSkipsAndReportsBadFiles(t *testing.T) {
 			t.Errorf("error %d = %v, want it to name %s", i, errs[i], name)
 		}
 	}
-	if got := provider.LocalErrors(); len(got) != len(wantErrs) {
+	if got := engine.LocalErrors(); len(got) != len(wantErrs) {
 		t.Errorf("LocalErrors = %v", got)
 	}
 	for _, id := range []string{"toy", "toyp"} {
-		e, ok := provider.Default().Get(id)
+		e, ok := engine.Default().Get(id)
 		if !ok || !strings.HasPrefix(e.Source, dir) {
 			t.Errorf("%s = %+v, %v; a good file after a bad one must still load", id, e, ok)
 		}
 	}
-	if e, _ := provider.Default().Get("vllm"); e.Source != "" {
+	if e, _ := engine.Default().Get("vllm"); e.Source != "" {
 		t.Errorf("vllm comes from %q; a local file redefined it", e.Source)
 	}
-	if _, err := provider.LookupEngine("toyp"); err != nil {
+	if _, err := engine.LookupEngine("toyp"); err != nil {
 		t.Errorf("a local engine cannot be selected: %v", err)
 	}
 
-	if errs := provider.UseLocal(filepath.Join(dir, "missing")); len(errs) != 0 {
+	if errs := engine.UseLocal(filepath.Join(dir, "missing")); len(errs) != 0 {
 		t.Errorf("a missing directory is an error: %v", errs)
 	}
-	if _, ok := provider.Default().Get("toy"); ok {
+	if _, ok := engine.Default().Get("toy"); ok {
 		t.Error("UseLocal kept engines from a directory it no longer reads")
 	}
-	if len(provider.LocalErrors()) != 0 {
+	if len(engine.LocalErrors()) != 0 {
 		t.Error("UseLocal kept errors from a directory it no longer reads")
 	}
 }
diff --git a/internal/provider/local.go b/internal/engine/local.go
similarity index 99%
rename from internal/provider/local.go
rename to internal/engine/local.go
index 90aa00d..0d7edc8 100644
--- a/internal/provider/local.go
+++ b/internal/engine/local.go
@@ -1,4 +1,4 @@
-package provider
+package engine
 
 import (
 	"errors"
diff --git a/internal/provider/provider.go b/internal/engine/registry.go
similarity index 96%
rename from internal/provider/provider.go
rename to internal/engine/registry.go
index ab06625..5506a2b 100644
--- a/internal/provider/provider.go
+++ b/internal/engine/registry.go
@@ -1,6 +1,5 @@
-// Package provider describes the inference servers a profile or recipe can
-// run on. Each one is an engine, defined as data in its own file under
-// catalogue/engines/.
+// Package engine describes the inference servers a profile or recipe can
+// run on, each defined as data in its own file under catalogue/engines/.
 //
 // An engine only decides how a profile becomes a command line: which
 // executable, which flags, which environment, and what that server can be
@@ -13,7 +12,7 @@
 // The data is parsed strictly. An unknown key, a duplicate id, or a
 // placeholder a template cannot fill is an error, so a typo fails when the
 // definition loads rather than reaching the server as an empty argument.
-package provider
+package engine
 
 import (
 	"fmt"
diff --git a/internal/provider/provider_test.go b/internal/engine/registry_test.go
similarity index 81%
rename from internal/provider/provider_test.go
rename to internal/engine/registry_test.go
index caf0de8..324d7b8 100644
--- a/internal/provider/provider_test.go
+++ b/internal/engine/registry_test.go
@@ -1,4 +1,4 @@
-package provider_test
+package engine_test
 
 import (
 	"os"
@@ -7,7 +7,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/engine"
 )
 
 // Every install from before providers existed has no PROVIDER line, and must
@@ -15,11 +15,11 @@ import (
 func TestLookupDefaultsToVLLM(t *testing.T) {
 	t.Parallel()
 	for _, name := range []string{"", "  ", "vllm", "VLLM"} {
-		p, err := provider.Lookup(name)
+		p, err := engine.Lookup(name)
 		if err != nil {
 			t.Fatalf("Lookup(%q): %v", name, err)
 		}
-		if p.Name() != provider.VLLM {
+		if p.Name() != engine.VLLM {
 			t.Errorf("Lookup(%q) = %s, want vllm", name, p.Name())
 		}
 	}
@@ -27,7 +27,7 @@ func TestLookupDefaultsToVLLM(t *testing.T) {
 
 func TestLookupRejectsUnknownNamingTheChoices(t *testing.T) {
 	t.Parallel()
-	_, err := provider.Lookup("ollama")
+	_, err := engine.Lookup("ollama")
 	if err == nil {
 		t.Fatal("an unknown provider must be an error, not a silent fallback to vLLM")
 	}
@@ -41,8 +41,8 @@ func TestLookupRejectsUnknownNamingTheChoices(t *testing.T) {
 // The argv vLLM gets is pinned, so moving it behind the seam changes nothing.
 func TestVLLMCommand(t *testing.T) {
 	t.Parallel()
-	p, _ := provider.Lookup(provider.VLLM)
-	got, err := p.Command(provider.Spec{
+	p, _ := engine.Lookup(engine.VLLM)
+	got, err := p.Command(engine.Spec{
 		Bin: "/opt/vllm/bin/vllm", Model: "Qwen/Qwen3-32B", Port: "8001",
 		MaxModelLen: "32768", MemUtilization: "0.85", APIKey: "sk-x",
 		KVCacheBytes: "1000", ExtraArgs: `--tool-call-parser hermes --chat-template "/a b.jinja"`,
@@ -63,8 +63,8 @@ func TestVLLMCommand(t *testing.T) {
 // is an argparse error that crash-loops the service.
 func TestMLXCommandCarriesNoVLLMFlags(t *testing.T) {
 	t.Parallel()
-	p, _ := provider.Lookup(provider.MLX)
-	got, err := p.Command(provider.Spec{
+	p, _ := engine.Lookup(engine.MLX)
+	got, err := p.Command(engine.Spec{
 		Bin: "/v/bin/mlx_lm.server", Model: "mlx-community/Qwen3-4B-4bit", Port: "8001",
 		MaxModelLen: "32768", MemUtilization: "0.85", APIKey: "sk-x", KVCacheBytes: "1000",
 		ExtraArgs: "--host 0.0.0.0 --max-tokens 8192",
@@ -88,7 +88,7 @@ func TestMLXCommandCarriesNoVLLMFlags(t *testing.T) {
 // and status look for.
 func TestLlamaCPPCommand(t *testing.T) {
 	t.Parallel()
-	p, _ := provider.Lookup(provider.LlamaCPP)
+	p, _ := engine.Lookup(engine.LlamaCPP)
 	bin := "/opt/homebrew/bin/llama-server"
 	for _, tc := range []struct {
 		name, model, apiKey, source string
@@ -100,7 +100,7 @@ func TestLlamaCPPCommand(t *testing.T) {
 		{"relative file", "qwen3-32b-q8_0.GGUF", "sk-x", "-m", true},
 		{"no key", "Qwen/Qwen2.5-0.5B-Instruct-GGUF", "", "-hf", false},
 	} {
-		got, err := p.Command(provider.Spec{
+		got, err := p.Command(engine.Spec{
 			Bin: bin, Model: tc.model, Port: "8001", APIKey: tc.apiKey,
 			MaxModelLen: "32768", MemUtilization: "0.85", KVCacheBytes: "1000",
 			ExtraArgs: `-c 65536 --chat-template-file "/a b.jinja"`,
@@ -119,7 +119,7 @@ func TestLlamaCPPCommand(t *testing.T) {
 		if !reflect.DeepEqual(got, want) {
 			t.Errorf("%s: argv\n got %q\nwant %q", tc.name, got, want)
 		}
-		envp, ok := p.(provider.EnvProvider)
+		envp, ok := p.(engine.EnvProvider)
 		if !ok {
 			t.Fatalf("llamacpp does not pass its key through the environment")
 		}
@@ -127,7 +127,7 @@ func TestLlamaCPPCommand(t *testing.T) {
 		if tc.wantKey {
 			wantEnv = []string{"LLAMA_API_KEY=" + tc.apiKey}
 		}
-		if gotEnv := envp.Env(provider.Spec{APIKey: tc.apiKey}); !reflect.DeepEqual(gotEnv, wantEnv) {
+		if gotEnv := envp.Env(engine.Spec{APIKey: tc.apiKey}); !reflect.DeepEqual(gotEnv, wantEnv) {
 			t.Errorf("%s: env = %q, want %q", tc.name, gotEnv, wantEnv)
 		}
 	}
@@ -138,9 +138,9 @@ func TestLlamaCPPCommand(t *testing.T) {
 
 func TestCommandRejectsUnparseableExtraArgs(t *testing.T) {
 	t.Parallel()
-	for _, name := range provider.Names() {
-		p, _ := provider.Lookup(name)
-		if _, err := p.Command(provider.Spec{Bin: "/x", Model: "m", Port: "1", ExtraArgs: `--a "unterminated`}); err == nil {
+	for _, name := range engine.Names() {
+		p, _ := engine.Lookup(name)
+		if _, err := p.Command(engine.Spec{Bin: "/x", Model: "m", Port: "1", ExtraArgs: `--a "unterminated`}); err == nil {
 			t.Errorf("%s: unbalanced quote in EXTRA_ARGS was accepted", name)
 		}
 	}
@@ -150,8 +150,8 @@ func TestCommandRejectsUnparseableExtraArgs(t *testing.T) {
 // told to fix VLLM_BIN.
 func TestValidateNamesTheProvidersBinKey(t *testing.T) {
 	t.Parallel()
-	p, _ := provider.Lookup(provider.MLX)
-	err := provider.Validate(p, provider.Spec{Model: "m", Port: "1"})
+	p, _ := engine.Lookup(engine.MLX)
+	err := engine.Validate(p, engine.Spec{Model: "m", Port: "1"})
 	if err == nil || !strings.Contains(err.Error(), "MLX_LM_BIN") {
 		t.Fatalf("Validate with no bin = %v, want it to name MLX_LM_BIN", err)
 	}
@@ -160,7 +160,7 @@ func TestValidateNamesTheProvidersBinKey(t *testing.T) {
 	if err := os.WriteFile(notExec, nil, 0o644); err != nil {
 		t.Fatal(err)
 	}
-	if err := provider.Validate(p, provider.Spec{Bin: notExec, Model: "m", Port: "1"}); err == nil {
+	if err := engine.Validate(p, engine.Spec{Bin: notExec, Model: "m", Port: "1"}); err == nil {
 		t.Fatal("a non-executable bin passed validation")
 	}
 
@@ -168,10 +168,10 @@ func TestValidateNamesTheProvidersBinKey(t *testing.T) {
 	if err := os.WriteFile(bin, nil, 0o755); err != nil {
 		t.Fatal(err)
 	}
-	if err := provider.Validate(p, provider.Spec{Bin: bin, Port: "1"}); err == nil || !strings.Contains(err.Error(), "MODEL") {
+	if err := engine.Validate(p, engine.Spec{Bin: bin, Port: "1"}); err == nil || !strings.Contains(err.Error(), "MODEL") {
 		t.Fatalf("Validate with no model = %v", err)
 	}
-	if err := provider.Validate(p, provider.Spec{Bin: bin, Model: "m", Port: "1"}); err != nil {
+	if err := engine.Validate(p, engine.Spec{Bin: bin, Model: "m", Port: "1"}); err != nil {
 		t.Fatalf("a complete spec failed validation: %v", err)
 	}
 }
diff --git a/internal/hardware/hardware.go b/internal/hardware/hardware.go
index 2e2ab93..b8ee8ba 100644
--- a/internal/hardware/hardware.go
+++ b/internal/hardware/hardware.go
@@ -31,7 +31,7 @@ import (
 	"slices"
 	"strings"
 
-	"github.com/coder/vllm-helper/catalogue"
+	"github.com/coder/llm-helper/catalogue"
 )
 
 // SchemaVersion is the definitions format this build reads. A file written
diff --git a/internal/hardware/hardware_test.go b/internal/hardware/hardware_test.go
index fd1bbff..8192dc1 100644
--- a/internal/hardware/hardware_test.go
+++ b/internal/hardware/hardware_test.go
@@ -8,7 +8,7 @@ import (
 	"testing"
 	"testing/fstest"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // The definitions replaced one hardcoded pattern. Until a definition is added
diff --git a/internal/hf/snippet.go b/internal/hf/snippet.go
index ea41475..5080df1 100644
--- a/internal/hf/snippet.go
+++ b/internal/hf/snippet.go
@@ -4,7 +4,7 @@ import (
 	"regexp"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // ownedFlags are set by vllm-helper itself from config and the profile, so a
diff --git a/internal/hf/testsupport.go b/internal/hf/testsupport.go
index aee6373..57aa1a2 100644
--- a/internal/hf/testsupport.go
+++ b/internal/hf/testsupport.go
@@ -3,7 +3,7 @@ package hf
 // SetBaseURLForTest points the Hub client at a test server and returns a
 // function that restores the previous value.
 //
-// Exported because catalog's tests need it as well: the judgements worth
+// Exported because resolve's tests need it as well: the judgements worth
 // testing -- which quants are offered, what is refused -- live there, and
 // they need a Hub that answers predictably. It is inert outside a test, since
 // nothing else ever calls it.
diff --git a/internal/nodes/registry.go b/internal/nodes/registry.go
index 575dda1..b81ae9a 100644
--- a/internal/nodes/registry.go
+++ b/internal/nodes/registry.go
@@ -8,7 +8,7 @@ import (
 	"sort"
 	"sync"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // NIC describes a network interface.
diff --git a/internal/plugin/coder/adopt_test.go b/internal/plugin/coder/adopt_test.go
index 16e0794..637a6df 100644
--- a/internal/plugin/coder/adopt_test.go
+++ b/internal/plugin/coder/adopt_test.go
@@ -7,7 +7,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/plugin/coder"
+	"github.com/coder/llm-helper/internal/plugin/coder"
 )
 
 // listing serves a fixed collection to every path, the way newClient does.
diff --git a/internal/plugin/coder/client_test.go b/internal/plugin/coder/client_test.go
index 9ed86af..f8546b5 100644
--- a/internal/plugin/coder/client_test.go
+++ b/internal/plugin/coder/client_test.go
@@ -7,7 +7,7 @@ import (
 	"net/http/httptest"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/plugin/coder"
+	"github.com/coder/llm-helper/internal/plugin/coder"
 )
 
 // Where a current deployment keeps the collection. The stand-in below answers
diff --git a/internal/plugin/coder/options_internal_test.go b/internal/plugin/coder/options_internal_test.go
index ab2e125..b013b84 100644
--- a/internal/plugin/coder/options_internal_test.go
+++ b/internal/plugin/coder/options_internal_test.go
@@ -6,7 +6,7 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 func cfgWith(kv map[string]string) *config.Config {
diff --git a/internal/plugin/coder/prune_test.go b/internal/plugin/coder/prune_test.go
index 503f90c..965f12c 100644
--- a/internal/plugin/coder/prune_test.go
+++ b/internal/plugin/coder/prune_test.go
@@ -7,7 +7,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/plugin/coder"
+	"github.com/coder/llm-helper/internal/plugin/coder"
 )
 
 // The collection is deployment-wide: an Anthropic model sits in the same list
diff --git a/internal/plugin/coder/route_test.go b/internal/plugin/coder/route_test.go
index c31156d..9082c9b 100644
--- a/internal/plugin/coder/route_test.go
+++ b/internal/plugin/coder/route_test.go
@@ -9,7 +9,7 @@ import (
 	"sync"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/plugin/coder"
+	"github.com/coder/llm-helper/internal/plugin/coder"
 )
 
 const (
diff --git a/internal/plugin/coder/sync.go b/internal/plugin/coder/sync.go
index 58a1d00..5edfc34 100644
--- a/internal/plugin/coder/sync.go
+++ b/internal/plugin/coder/sync.go
@@ -12,10 +12,10 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // Configuration keys. All three are required: a URL without a token, or
@@ -356,7 +356,7 @@ func modelConfig(cfg *config.Config, p *profile.Profile, enabled bool) (ModelCon
 		// failing sync for that would leave the previous model enabled after a
 		// switch. An unknown PROVIDER gets no such grace: it cannot be served,
 		// and inventing a context limit would hide that.
-		if prov, err := provider.Lookup(p.Provider()); err == nil && !prov.Capabilities().VLLMArgs {
+		if prov, err := engine.Lookup(p.Provider()); err == nil && !prov.Capabilities().VLLMArgs {
 			raw = strconv.Itoa(profile.DefaultMaxModelLen)
 		}
 	}
diff --git a/internal/plugin/coder/sync_internal_test.go b/internal/plugin/coder/sync_internal_test.go
index af4f054..733ff00 100644
--- a/internal/plugin/coder/sync_internal_test.go
+++ b/internal/plugin/coder/sync_internal_test.go
@@ -10,9 +10,9 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func profileWith(t *testing.T, body string) *profile.Profile {
diff --git a/internal/plugin/plugin.go b/internal/plugin/plugin.go
index 4086579..12ee993 100644
--- a/internal/plugin/plugin.go
+++ b/internal/plugin/plugin.go
@@ -18,8 +18,8 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // Event is a point in a profile's life that a plugin can react to.
diff --git a/internal/plugin/registry.go b/internal/plugin/registry.go
index 4cbeba3..d72efc3 100644
--- a/internal/plugin/registry.go
+++ b/internal/plugin/registry.go
@@ -9,7 +9,7 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // registered is the built-in set, populated by each plugin's init. Plugins
diff --git a/internal/plugin/registry_internal_test.go b/internal/plugin/registry_internal_test.go
index d28cef8..d640227 100644
--- a/internal/plugin/registry_internal_test.go
+++ b/internal/plugin/registry_internal_test.go
@@ -8,7 +8,7 @@ import (
 
 	"github.com/spf13/cobra"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 type fake struct {
diff --git a/internal/plugin/spark/spark.go b/internal/plugin/spark/spark.go
index ea2b63b..55f8c20 100644
--- a/internal/plugin/spark/spark.go
+++ b/internal/plugin/spark/spark.go
@@ -14,10 +14,10 @@ import (
 	"github.com/spf13/cobra"
 	"golang.org/x/term"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/plugin"
-	"github.com/coder/vllm-helper/internal/toolchain"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/plugin"
+	"github.com/coder/llm-helper/internal/toolchain"
 )
 
 func init() { plugin.Register(Plugin{}) }
diff --git a/internal/profile/generated_test.go b/internal/profile/generated_test.go
index aaafed7..4f12adf 100644
--- a/internal/profile/generated_test.go
+++ b/internal/profile/generated_test.go
@@ -5,8 +5,8 @@ import (
 	"path/filepath"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func loadFrom(t *testing.T, body string) *profile.Profile {
diff --git a/internal/profile/profile.go b/internal/profile/profile.go
index 3ba6e60..a547eef 100644
--- a/internal/profile/profile.go
+++ b/internal/profile/profile.go
@@ -12,8 +12,8 @@ import (
 	"strings"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
 )
 
 type Profile struct {
diff --git a/internal/profile/profile_test.go b/internal/profile/profile_test.go
index c811c6c..98378c7 100644
--- a/internal/profile/profile_test.go
+++ b/internal/profile/profile_test.go
@@ -6,9 +6,9 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 func newDirs(t *testing.T) config.Dirs {
diff --git a/internal/recipe/args.go b/internal/recipe/args.go
index 1a8fb49..b695960 100644
--- a/internal/recipe/args.go
+++ b/internal/recipe/args.go
@@ -4,7 +4,7 @@ import (
 	"fmt"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // Target is the machine a recipe is resolved for. The hardware definitions
diff --git a/internal/recipe/container.go b/internal/recipe/container.go
index f5b343b..1c456a0 100644
--- a/internal/recipe/container.go
+++ b/internal/recipe/container.go
@@ -5,7 +5,7 @@ import (
 	"sort"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/engine"
 )
 
 // ContainerSpec is everything needed to start one node of a recipe. RankSpec
@@ -20,7 +20,7 @@ type ContainerSpec struct {
 	// Mounts are host directories, as absolute paths, bind-mounted rather
 	// than baked in so a re-launch does not re-download weights or re-JIT
 	// kernels, which is most of the startup cost.
-	Mounts []provider.Mount
+	Mounts []engine.Mount
 	Env    map[string]string
 	Args   []string
 	// APIKey is kept out of Env so it can be redacted when the command is
diff --git a/internal/recipe/launch.go b/internal/recipe/launch.go
index 9ce1411..421b80e 100644
--- a/internal/recipe/launch.go
+++ b/internal/recipe/launch.go
@@ -7,8 +7,8 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // Placeholders a recipe may reference in its env values. They are discovered on
@@ -73,7 +73,7 @@ func placeholderName(v string) (string, bool) {
 // overrides are merged in, so an override is never expanded.
 func (r TestedRecipe) ServeArgs() []string {
 	e := r.Engine()
-	v := provider.RoleValues{Model: r.Model, Port: r.Port, Nodes: r.Nodes, Params: r.Params, Models: r.ModelRefs()}
+	v := engine.RoleValues{Model: r.Model, Port: r.Port, Nodes: r.Nodes, Params: r.Params, Models: r.ModelRefs()}
 	args := e.ExpandArgs(r.Args, v)
 	// Host and port are the launcher's business, not the model's, which is why
 	// the recipe carries "port" as a field instead of inside args. The engine
@@ -220,7 +220,7 @@ func (r TestedRecipe) RankSpec(k Rank) (ContainerSpec, error) {
 	if err != nil {
 		return ContainerSpec{}, err
 	}
-	entry, args, env := e.RenderRole(role, provider.RoleValues{
+	entry, args, env := e.RenderRole(role, engine.RoleValues{
 		Model: r.Model, Port: r.Port, Rank: k.Rank, Nodes: r.Nodes,
 		MasterAddr: k.MasterAddr, MasterPort: k.MasterPort,
 		Args: k.ServeArgs, Params: r.Params, Models: r.ModelRefs(),
@@ -296,7 +296,7 @@ func (r TestedRecipe) StagingScript() string {
 	var b strings.Builder
 	b.WriteString(stagedFunc)
 	for _, m := range r.Models {
-		dir := provider.SnapshotPath(cache, provider.ModelRef{ID: m.ID, Revision: m.Revision})
+		dir := engine.SnapshotPath(cache, engine.ModelRef{ID: m.ID, Revision: m.Revision})
 		fmt.Fprintf(&b, "staged %s || echo %s\n", ShellQuote(dir), ShellQuote(m.ID+"@"+m.Revision))
 	}
 	return b.String()
diff --git a/internal/recipe/local.go b/internal/recipe/local.go
index 8d2f101..b085362 100644
--- a/internal/recipe/local.go
+++ b/internal/recipe/local.go
@@ -29,7 +29,7 @@ var (
 // launches. A missing dir is not an error; a bad file is skipped and
 // reported, and the files after it still load.
 //
-// Call it after provider.UseLocal, since a local recipe may run on a local
+// Call it after engine.UseLocal, since a local recipe may run on a local
 // engine.
 func UseLocal(dir string) []error {
 	var errs []error
diff --git a/internal/recipe/tested.go b/internal/recipe/tested.go
index 01447af..784d914 100644
--- a/internal/recipe/tested.go
+++ b/internal/recipe/tested.go
@@ -13,9 +13,9 @@ import (
 	"sort"
 	"strings"
 
-	"github.com/coder/vllm-helper/catalogue"
-	"github.com/coder/vllm-helper/internal/hardware"
-	"github.com/coder/vllm-helper/internal/provider"
+	"github.com/coder/llm-helper/catalogue"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 // SchemaVersion is the recipe format this build reads, in a file under
@@ -299,14 +299,14 @@ func (r TestedRecipe) validate() error {
 			add("requires %q names no hardware definition", r.Requires)
 		}
 	}
-	e, err := provider.LookupEngine(r.Provider)
+	e, err := engine.LookupEngine(r.Provider)
 	if err != nil {
 		add("%v", err)
-	} else if !e.Has(provider.RuntimeContainer) {
+	} else if !e.Has(engine.RuntimeContainer) {
 		add("engine %s has no container runtime, so it cannot launch a recipe", e.ID)
 	} else {
 		for name := range r.Params {
-			if !slices.ContainsFunc(e.Params, func(p provider.Param) bool { return p.Name == name }) {
+			if !slices.ContainsFunc(e.Params, func(p engine.Param) bool { return p.Name == name }) {
 				add("params sets %q, which engine %s does not declare", name, e.ID)
 			}
 		}
@@ -317,7 +317,7 @@ func (r TestedRecipe) validate() error {
 		for _, m := range r.Models {
 			have[m.Role] = true
 		}
-		for _, role := range append(e.UsesModelPaths(), provider.ModelRolesIn(r.Args)...) {
+		for _, role := range append(e.UsesModelPaths(), engine.ModelRolesIn(r.Args)...) {
 			if !have[role] {
 				add("engine %s needs a pinned %s model, and models has none", e.ID, role)
 			}
@@ -333,8 +333,8 @@ func (r TestedRecipe) validate() error {
 			add("model %q is not an org/name repo id", m.ID)
 		case !revisionPattern.MatchString(m.Revision):
 			add("model %s: revision %q is not a full commit sha", m.ID, m.Revision)
-		case !slices.Contains(provider.ModelRoles, m.Role):
-			add("model %s: role %q must be one of %s", m.ID, m.Role, strings.Join(provider.ModelRoles, ", "))
+		case !slices.Contains(engine.ModelRoles, m.Role):
+			add("model %s: role %q must be one of %s", m.ID, m.Role, strings.Join(engine.ModelRoles, ", "))
 		case roles[m.Role]:
 			add("model %s: a recipe pins one model per role, and %s is taken", m.ID, m.Role)
 		case m.Staging != "download":
@@ -351,16 +351,16 @@ func (r TestedRecipe) validate() error {
 // Engine is the engine that runs the recipe. Recipes are validated when they
 // load, so an unknown engine cannot get this far; the zero Engine it would
 // return renders nothing.
-func (r TestedRecipe) Engine() provider.Engine {
-	e, _ := provider.LookupEngine(r.Provider)
+func (r TestedRecipe) Engine() engine.Engine {
+	e, _ := engine.LookupEngine(r.Provider)
 	return e
 }
 
 // ModelRefs are the recipe's pinned models by role.
-func (r TestedRecipe) ModelRefs() map[string]provider.ModelRef {
-	out := map[string]provider.ModelRef{}
+func (r TestedRecipe) ModelRefs() map[string]engine.ModelRef {
+	out := map[string]engine.ModelRef{}
 	for _, m := range r.Models {
-		out[m.Role] = provider.ModelRef{ID: m.ID, Revision: m.Revision}
+		out[m.Role] = engine.ModelRef{ID: m.ID, Revision: m.Revision}
 	}
 	return out
 }
diff --git a/internal/recipe/tested_internal_test.go b/internal/recipe/tested_internal_test.go
index 65e8f00..ff5b6b2 100644
--- a/internal/recipe/tested_internal_test.go
+++ b/internal/recipe/tested_internal_test.go
@@ -6,7 +6,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/hardware"
+	"github.com/coder/llm-helper/internal/hardware"
 )
 
 func TestTestedRecipesAreWellFormed(t *testing.T) {
diff --git a/internal/recipe/validate_internal_test.go b/internal/recipe/validate_internal_test.go
index 3e3c21c..e9349c3 100644
--- a/internal/recipe/validate_internal_test.go
+++ b/internal/recipe/validate_internal_test.go
@@ -9,7 +9,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/catalogue"
+	"github.com/coder/llm-helper/catalogue"
 )
 
 func glm(t *testing.T) TestedRecipe {
diff --git a/internal/catalog/catalog.go b/internal/resolve/resolve.go
similarity index 97%
rename from internal/catalog/catalog.go
rename to internal/resolve/resolve.go
index 1d56d1a..4777f4a 100644
--- a/internal/catalog/catalog.go
+++ b/internal/resolve/resolve.go
@@ -1,4 +1,4 @@
-// Package catalog turns a HuggingFace repo into something this machine can
+// Package resolve turns a HuggingFace repo into something this machine can
 // serve: what a model would cost here, and the settings to serve it with.
 //
 // It sits under both the CLI and the web panel. Both need the same judgements
@@ -6,7 +6,7 @@
 // -- and those judgements have a safe direction that is easy to lose track of.
 // Keeping one copy is what stops the two interfaces from disagreeing about
 // whether a 200 GB model fits in 84 GB.
-package catalog
+package resolve
 
 import (
 	"context"
@@ -18,12 +18,12 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // repoPattern is the shape of a HuggingFace repo id: one owner, one name.
@@ -260,7 +260,7 @@ func cardFlags(flags hf.DetectedFlags, caps vllm.Caps) (args, note string) {
 // so search and the panel never offer what the server cannot load. The card's
 // flags are vllm serve's and this server is not taking them, so they stop
 // reading as detected -- a pull would not have copied them either.
-func ForProvider(d Detail, prov provider.Provider) Detail {
+func ForProvider(d Detail, prov engine.Provider) Detail {
 	caps := prov.Capabilities()
 	if caps.VLLMArgs {
 		return d
@@ -308,7 +308,7 @@ type Plan struct {
 // Everything PrepareProfile works out (card flags, parser checks, a sized KV
 // cache, the template patch) is in vLLM's terms, so a profile for any other
 // provider names the model and leaves the rest to that server's defaults.
-func PrepareFor(ctx context.Context, dirs config.Dirs, prov provider.Provider, repo string, choice *Choice, caps vllm.Caps, contextTarget int) Plan {
+func PrepareFor(ctx context.Context, dirs config.Dirs, prov engine.Provider, repo string, choice *Choice, caps vllm.Caps, contextTarget int) Plan {
 	if prov.Capabilities().VLLMArgs {
 		return PrepareProfile(ctx, dirs, repo, choice, caps, contextTarget)
 	}
diff --git a/internal/catalog/catalog_internal_test.go b/internal/resolve/resolve_internal_test.go
similarity index 96%
rename from internal/catalog/catalog_internal_test.go
rename to internal/resolve/resolve_internal_test.go
index 4e059cb..bdd8fb3 100644
--- a/internal/catalog/catalog_internal_test.go
+++ b/internal/resolve/resolve_internal_test.go
@@ -1,4 +1,4 @@
-package catalog
+package resolve
 
 import (
 	"context"
@@ -11,12 +11,12 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // A repo id becomes both a URL sent to the Hub and, through NameFromRepo, part
@@ -143,11 +143,11 @@ func TestPrepareForAnotherProviderNamesOnlyTheModel(t *testing.T) {
 	defer srv.Close()
 	defer hf.SetBaseURLForTest(srv.URL)()
 
-	mlx, _ := provider.Lookup(provider.MLX)
+	mlx, _ := engine.Lookup(engine.MLX)
 	p := PrepareFor(context.Background(), config.Dirs{Data: t.TempDir()}, mlx,
 		"mlx-community/Qwen3-4B-4bit", nil, vllm.Caps{}, 0)
 
-	if p.Opts.Provider != provider.MLX || p.Opts.Repo != "mlx-community/Qwen3-4B-4bit" {
+	if p.Opts.Provider != engine.MLX || p.Opts.Repo != "mlx-community/Qwen3-4B-4bit" {
 		t.Errorf("Opts = %+v, want the repo and PROVIDER=mlx", p.Opts)
 	}
 	if p.Opts.ExtraArgs != "" || p.Opts.KVCacheBytes != 0 || p.Opts.MaxModelLen != "" {
@@ -160,7 +160,7 @@ func TestPrepareForAnotherProviderNamesOnlyTheModel(t *testing.T) {
 
 	// vLLM still gets the full treatment, and a profile that names no
 	// provider, as every vLLM profile always has.
-	vl, _ := provider.Lookup(provider.VLLM)
+	vl, _ := engine.Lookup(engine.VLLM)
 	v := PrepareFor(context.Background(), config.Dirs{Data: t.TempDir()}, vl,
 		"Qwen/Qwen3-32B", nil, vllm.Caps{}, 0)
 	if v.Opts.Provider != "" {
@@ -200,14 +200,14 @@ func TestChooseGGUFQuantForLlamaCPP(t *testing.T) {
 	defer srv.Close()
 	defer hf.SetBaseURLForTest(srv.URL)()
 	ctx := context.Background()
-	llama, _ := provider.Lookup(provider.LlamaCPP)
+	llama, _ := engine.Lookup(engine.LlamaCPP)
 
 	choice, err := ChooseGGUFQuant(ctx, repo, "bf16")
 	if err != nil {
 		t.Fatalf("a split GGUF was refused: %v", err)
 	}
 	p := PrepareFor(ctx, config.Dirs{Data: t.TempDir()}, llama, repo, choice, vllm.Caps{}, 0)
-	if p.Opts.Repo != repo+":BF16" || p.Opts.Name != "qwen_qwen3-32b-gguf-bf16" || p.Opts.Provider != provider.LlamaCPP {
+	if p.Opts.Repo != repo+":BF16" || p.Opts.Name != "qwen_qwen3-32b-gguf-bf16" || p.Opts.Provider != engine.LlamaCPP {
 		t.Errorf("Opts = %+v, want MODEL %s:BF16 in profile qwen_qwen3-32b-gguf-bf16", p.Opts, repo)
 	}
 
@@ -238,21 +238,21 @@ func TestForProvider(t *testing.T) {
 			{Name: "Q4_K_M.gguf", Label: "Q4_K_M", SizeGB: 19, GGUF: true, Blocked: "a mixture-of-experts model"},
 		},
 	}
-	get := func(name string) provider.Provider {
-		p, err := provider.Lookup(name)
+	get := func(name string) engine.Provider {
+		p, err := engine.Lookup(name)
 		if err != nil {
 			t.Fatal(err)
 		}
 		return p
 	}
 
-	if got := ForProvider(inspected, get(provider.VLLM)); got.ArchBlocked == "" || got.GGUFReady || got.Best != nil {
+	if got := ForProvider(inspected, get(engine.VLLM)); got.ArchBlocked == "" || got.GGUFReady || got.Best != nil {
 		t.Errorf("vllm: %+v, want Inspect's view unchanged", got)
 	}
 	inspected.DetectedFlags = "--tokenizer Qwen/Qwen3-32B"
 	inspected.DroppedFlags = []string{"--a-flag-this-vllm-lacks"}
 
-	llama := ForProvider(inspected, get(provider.LlamaCPP))
+	llama := ForProvider(inspected, get(engine.LlamaCPP))
 	if !llama.GGUFReady || llama.ArchBlocked != "" {
 		t.Errorf("llamacpp: GGUFReady %v, ArchBlocked %q, want the plugin's limits lifted", llama.GGUFReady, llama.ArchBlocked)
 	}
@@ -271,7 +271,7 @@ func TestForProvider(t *testing.T) {
 		t.Error("ForProvider changed the Detail it was given")
 	}
 
-	mlx := ForProvider(inspected, get(provider.MLX))
+	mlx := ForProvider(inspected, get(engine.MLX))
 	if mlx.ArchBlocked == "" || mlx.Best != nil {
 		t.Errorf("mlx: ArchBlocked %q, Best %+v, want the GGUF repo refused", mlx.ArchBlocked, mlx.Best)
 	}
@@ -287,10 +287,10 @@ func TestForProvider(t *testing.T) {
 		{Name: ".", Label: "default", SizeGB: 65},
 		{Name: "m-Q8_0.gguf", Label: "Q8_0", SizeGB: 34, GGUF: true},
 	}}
-	if got := ForProvider(mixed, get(provider.LlamaCPP)); got.Best == nil || got.Best.Label != "Q8_0" || got.All[0].Servable() {
+	if got := ForProvider(mixed, get(engine.LlamaCPP)); got.Best == nil || got.Best.Label != "Q8_0" || got.All[0].Servable() {
 		t.Errorf("llamacpp mixed repo: Best %+v, safetensors blocked %q, want Q8_0 and the safetensors ruled out", got.Best, got.All[0].Blocked)
 	}
-	if got := ForProvider(mixed, get(provider.MLX)); got.Best == nil || got.Best.Label != "default" || got.All[1].Servable() {
+	if got := ForProvider(mixed, get(engine.MLX)); got.Best == nil || got.Best.Label != "default" || got.All[1].Servable() {
 		t.Errorf("mlx mixed repo: Best %+v, want the safetensors and the GGUF ruled out", got.Best)
 	}
 }
diff --git a/internal/catalog/tiktoken.go b/internal/resolve/tiktoken.go
similarity index 98%
rename from internal/catalog/tiktoken.go
rename to internal/resolve/tiktoken.go
index f6513fa..0f5f413 100644
--- a/internal/catalog/tiktoken.go
+++ b/internal/resolve/tiktoken.go
@@ -1,4 +1,4 @@
-package catalog
+package resolve
 
 import (
 	"fmt"
@@ -9,8 +9,8 @@ import (
 	"strings"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // harmonyEncodings are the vocab files the harmony tokenizer reads, by the
diff --git a/internal/catalog/tiktoken_internal_test.go b/internal/resolve/tiktoken_internal_test.go
similarity index 98%
rename from internal/catalog/tiktoken_internal_test.go
rename to internal/resolve/tiktoken_internal_test.go
index 5ea5a03..1d68b90 100644
--- a/internal/catalog/tiktoken_internal_test.go
+++ b/internal/resolve/tiktoken_internal_test.go
@@ -1,4 +1,4 @@
-package catalog
+package resolve
 
 import (
 	"net/http"
@@ -8,7 +8,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // setup creates the encodings directory and points vLLM at it, but nothing
diff --git a/internal/toolchain/toolchain_darwin.go b/internal/toolchain/toolchain_darwin.go
index b9aca79..a0fd6ce 100644
--- a/internal/toolchain/toolchain_darwin.go
+++ b/internal/toolchain/toolchain_darwin.go
@@ -1,6 +1,6 @@
 package toolchain
 
-import "github.com/coder/vllm-helper/internal/detect"
+import "github.com/coder/llm-helper/internal/detect"
 
 // Gather checks what a macOS source build of vLLM's CPU backend needs: it
 // has no Apple Silicon wheels, so it is compiled locally against Apple Clang
diff --git a/internal/vllm/validate.go b/internal/vllm/validate.go
index 8798d2f..4aae029 100644
--- a/internal/vllm/validate.go
+++ b/internal/vllm/validate.go
@@ -4,7 +4,7 @@ import (
 	"fmt"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/config"
 )
 
 // parserFlags are the flags whose values must name something in the
diff --git a/internal/web/activate.go b/internal/web/activate.go
index 10763fd..3f3bc45 100644
--- a/internal/web/activate.go
+++ b/internal/web/activate.go
@@ -10,8 +10,8 @@ import (
 	"path/filepath"
 	"strings"
 
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // activate switches this machine to a saved profile.
diff --git a/internal/web/api.go b/internal/web/api.go
index bcca212..80db119 100644
--- a/internal/web/api.go
+++ b/internal/web/api.go
@@ -11,9 +11,9 @@ import (
 	"strconv"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/profile"
 )
 
 // run executes a command and reports only whether it succeeded. Used for
diff --git a/internal/web/docs.go b/internal/web/docs.go
index 033361e..f4110f4 100644
--- a/internal/web/docs.go
+++ b/internal/web/docs.go
@@ -27,7 +27,7 @@ import (
 	"github.com/yuin/goldmark/text"
 	"github.com/yuin/goldmark/util"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 //go:embed docs.tmpl
diff --git a/internal/web/docs_internal_test.go b/internal/web/docs_internal_test.go
index ccc99aa..1b2067a 100644
--- a/internal/web/docs_internal_test.go
+++ b/internal/web/docs_internal_test.go
@@ -7,7 +7,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 func getDocs(t *testing.T, s *Server, path string) *httptest.ResponseRecorder {
diff --git a/internal/web/docssite.go b/internal/web/docssite.go
index 64428b3..fc81956 100644
--- a/internal/web/docssite.go
+++ b/internal/web/docssite.go
@@ -14,7 +14,7 @@ import (
 	"net/http"
 	"strings"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 // docsSiteSummary is the one-line description llms.txt leads with. It is the
diff --git a/internal/web/docssite_internal_test.go b/internal/web/docssite_internal_test.go
index cdcadfb..c68f463 100644
--- a/internal/web/docssite_internal_test.go
+++ b/internal/web/docssite_internal_test.go
@@ -9,7 +9,7 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/coder/vllm-helper/docs"
+	"github.com/coder/llm-helper/docs"
 )
 
 const testDocsBase = "https://docs.example.test"
diff --git a/internal/web/models.go b/internal/web/models.go
index 9e39c2c..ae68251 100644
--- a/internal/web/models.go
+++ b/internal/web/models.go
@@ -9,13 +9,13 @@ import (
 	"strings"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/catalog"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/detect"
-	"github.com/coder/vllm-helper/internal/hf"
-	"github.com/coder/vllm-helper/internal/profile"
-	"github.com/coder/vllm-helper/internal/provider"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/detect"
+	"github.com/coder/llm-helper/internal/engine"
+	"github.com/coder/llm-helper/internal/hf"
+	"github.com/coder/llm-helper/internal/profile"
+	"github.com/coder/llm-helper/internal/resolve"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // hubTimeout bounds a call that leaves this machine. The Hub client already
@@ -38,7 +38,7 @@ func repoParam(r *http.Request) (string, error) {
 	if repo == "" {
 		return "", errors.New("no model given")
 	}
-	if !catalog.ValidRepo(repo) {
+	if !resolve.ValidRepo(repo) {
 		return "", fmt.Errorf("%q is not a HuggingFace repo id -- expected something like org/model-name", repo)
 	}
 	return repo, nil
@@ -49,7 +49,7 @@ func (s *Server) budget() detect.WeightsBudget { return detect.ModelBudget(s.cfg
 
 type searchResponse struct {
 	Query   string           `json:"query"`
-	Results []catalog.Result `json:"results"`
+	Results []resolve.Result `json:"results"`
 }
 
 // handleSearch queries the Hub. Read-only and outbound: it tells the caller
@@ -69,9 +69,9 @@ func (s *Server) handleSearch(w http.ResponseWriter, r *http.Request) {
 		})
 		return
 	}
-	out := make([]catalog.Result, 0, len(models))
+	out := make([]resolve.Result, 0, len(models))
 	for _, m := range models {
-		out = append(out, catalog.ToResult(m))
+		out = append(out, resolve.ToResult(m))
 	}
 	writeJSON(w, http.StatusOK, searchResponse{Query: query, Results: out})
 }
@@ -80,7 +80,7 @@ func (s *Server) handleSearch(w http.ResponseWriter, r *http.Request) {
 // this model already exists, so the UI can offer "switch to it" rather than
 // a pull that would fail as a duplicate.
 type inspectResponse struct {
-	catalog.Detail
+	resolve.Detail
 	ExistingProfile string `json:"existing_profile,omitempty"`
 }
 
@@ -108,14 +108,14 @@ func (s *Server) handleInspect(w http.ResponseWriter, r *http.Request) {
 	}
 	ctx, cancel := context.WithTimeout(r.Context(), hubTimeout)
 	defer cancel()
-	prov, err := provider.Lookup(s.cfg.Provider())
+	prov, err := engine.Lookup(s.cfg.Provider())
 	if err != nil {
 		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
 		return
 	}
 
 	type result struct {
-		d   catalog.Detail
+		d   resolve.Detail
 		err error
 	}
 	ch := make(chan result, 1)
@@ -131,7 +131,7 @@ func (s *Server) handleInspect(w http.ResponseWriter, r *http.Request) {
 		if prov.Capabilities().VLLMArgs {
 			caps = s.caps()
 		}
-		ch <- result{d: catalog.ForProvider(catalog.Inspect(ctx, m, s.budget(), caps, s.cfg.KVContextTarget()), prov)}
+		ch <- result{d: resolve.ForProvider(resolve.Inspect(ctx, m, s.budget(), caps, s.cfg.KVContextTarget()), prov)}
 	}()
 
 	select {
@@ -188,19 +188,19 @@ func (s *Server) handlePull(w http.ResponseWriter, r *http.Request) {
 	// An explicit quant is resolved against what the repo actually holds,
 	// so a stale page cannot ask for one that is gone -- or one that vLLM
 	// has no way to load.
-	prov, err := provider.Lookup(s.cfg.Provider())
+	prov, err := engine.Lookup(s.cfg.Provider())
 	if err != nil {
 		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
 		return
 	}
-	var choice *catalog.Choice
+	var choice *resolve.Choice
 	if want := strings.TrimSpace(r.URL.Query().Get("quant")); want != "" && !prov.Capabilities().VLLMArgs {
 		if !prov.Capabilities().GGUFQuant {
 			writeJSON(w, http.StatusBadRequest, map[string]string{
 				"error": "a quant picks one GGUF file out of a repo, but new profiles here are served by " + prov.Name() + ", which does not load GGUF"})
 			return
 		}
-		if choice, err = catalog.ChooseGGUFQuant(r.Context(), repo, want); err != nil {
+		if choice, err = resolve.ChooseGGUFQuant(r.Context(), repo, want); err != nil {
 			writeJSON(w, http.StatusBadRequest, map[string]string{"error": err.Error()})
 			return
 		}
@@ -211,8 +211,8 @@ func (s *Server) handlePull(w http.ResponseWriter, r *http.Request) {
 				"error": "could not re-read " + repo + " from HuggingFace: " + err.Error()})
 			return
 		}
-		d := catalog.Inspect(r.Context(), m, s.budget(), s.caps(), s.cfg.KVContextTarget())
-		v, ok := catalog.FindVariant(d.All, want)
+		d := resolve.Inspect(r.Context(), m, s.budget(), s.caps(), s.cfg.KVContextTarget())
+		v, ok := resolve.FindVariant(d.All, want)
 		switch {
 		case !ok:
 			writeJSON(w, http.StatusBadRequest, map[string]string{
@@ -231,7 +231,7 @@ func (s *Server) handlePull(w http.ResponseWriter, r *http.Request) {
 				"error": "this vLLM cannot load GGUF -- install vllm-gguf-plugin first"})
 			return
 		}
-		choice = &catalog.Choice{Variant: v, BaseModel: d.BaseModel}
+		choice = &resolve.Choice{Variant: v, BaseModel: d.BaseModel}
 	}
 
 	name := profile.NameFromRepo(repo)
@@ -270,7 +270,7 @@ func (s *Server) handlePull(w http.ResponseWriter, r *http.Request) {
 		if prov.Capabilities().VLLMArgs {
 			caps = s.caps()
 		}
-		plan := catalog.PrepareFor(r.Context(), s.cfg.Dirs, prov, repo, choice, caps, s.cfg.KVContextTarget())
+		plan := resolve.PrepareFor(r.Context(), s.cfg.Dirs, prov, repo, choice, caps, s.cfg.KVContextTarget())
 		out.Notes = append(out.Notes, plan.Notes...)
 		if _, err := profile.Create(s.cfg.Dirs, plan.Opts); err != nil {
 			writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
diff --git a/internal/web/server.go b/internal/web/server.go
index 51c0874..951e5db 100644
--- a/internal/web/server.go
+++ b/internal/web/server.go
@@ -25,10 +25,10 @@ import (
 	"sync"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/service"
-	"github.com/coder/vllm-helper/internal/vllm"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/service"
+	"github.com/coder/llm-helper/internal/vllm"
 )
 
 // Server holds everything the handlers need.
diff --git a/internal/web/server_internal_test.go b/internal/web/server_internal_test.go
index 420d92c..fa3dd39 100644
--- a/internal/web/server_internal_test.go
+++ b/internal/web/server_internal_test.go
@@ -14,9 +14,9 @@ import (
 	"testing"
 	"time"
 
-	"github.com/coder/vllm-helper/internal/config"
-	"github.com/coder/vllm-helper/internal/control"
-	"github.com/coder/vllm-helper/internal/service"
+	"github.com/coder/llm-helper/internal/config"
+	"github.com/coder/llm-helper/internal/control"
+	"github.com/coder/llm-helper/internal/service"
 )
 
 // newTestServer builds a server over a throwaway config tree that contains
diff --git a/main.go b/main.go
index f984cb6..3f6017c 100644
--- a/main.go
+++ b/main.go
@@ -7,7 +7,7 @@ package main
 import (
 	"os"
 
-	"github.com/coder/vllm-helper/internal/cli"
+	"github.com/coder/llm-helper/internal/cli"
 )
 
 func main() {
diff --git a/scripts/gendocs/main.go b/scripts/gendocs/main.go
index 13eed89..de2ccc6 100644
--- a/scripts/gendocs/main.go
+++ b/scripts/gendocs/main.go
@@ -15,8 +15,8 @@ import (
 	"regexp"
 	"sort"
 
-	"github.com/coder/vllm-helper/docs"
-	"github.com/coder/vllm-helper/internal/cli"
+	"github.com/coder/llm-helper/docs"
+	"github.com/coder/llm-helper/internal/cli"
 )
 
 func main() {
diff --git a/tools/recipediff/go.mod b/tools/recipediff/go.mod
index f7bb329..ed68cfd 100644
--- a/tools/recipediff/go.mod
+++ b/tools/recipediff/go.mod
@@ -1,4 +1,4 @@
-module github.com/coder/vllm-helper/tools/recipediff
+module github.com/coder/llm-helper/tools/recipediff
 
 go 1.26.5