[fazzatti]

Convee can execute intentionally authored functions, steps and pipelines in dedicated workers while keeping the normal Step API and caller-owned plugin lifecycle. Each worker definition exposes one callable. Parallel groups have independent capacities, and nested split results retain their exact association and TypeScript types when joined.

// word-count.worker.ts
export default worker(countWords, { module: import.meta.url });

// caller
const processDocuments = parallel({
  workers: { words: { worker: wordCount.use(wordHooks), count: 3 } },
  split: ({ words }, texts: string[]) => texts.map((text) => words.job(text)),
  join: (counts) => counts,
});

Behavior

• worker(callable, { module: import.meta.url }) is a directly imported, immutable definition. worker.use() returns a configured copy with per-job caller plugins; embedded native plugins execute inside the worker. parallel.use() surrounds the whole invocation.
• Ordinary parallel is a reusable Step with invocation-owned workers and automatic cleanup on every path. persistentParallel owns .step, optional warmup, graceful close, terminal termination/disposal and await using; optional reopening and idle timeout control intentional reuse.
• Named groups schedule independently. Nested arrays/tuples preserve order and inference without task-name registries, roles, recipes, explicit pool generics or a separate public pool executor.
• Whole batches reserve one bounded capacity budget before job hooks. Slots remain reserved through caller finalization; persistent calls share capacity. Explicit job context is snapshotted and fresh per execution.
• all aggregates unrecovered errors after every admitted job settles; collect delivers the matching outcome tree. Remote diagnostics retain bounded Convee details, causes and group/job/path identities.
• Startup and dispatch deadlines are separate. Crashes/timeouts fail their job without replay, await termination and use a bounded, group-wide replacement budget for later work. Exhaustion closes only that group. Constructor cancellation, late replies, cleanup failures and close/reopen races have dedicated regressions.
• Built-in Deno, Node/Bun and browser adapters share the same public API. The optional /workers/vite integration preserves worker-file syntax and emits entries for actual development and production builds.

Validation

• Full deno task verify passes locally on Deno 2.9.7: 44 runtime groups / 823 BDD steps, public type contracts, 29 README examples, tooling, architecture, type-depth budgets, isolated source-package consumers and JSR dry run.
• Exact production coverage: 2834/2834 lines, 1093/1093 branches, 253/253 instrumented functions — 100% each, including the Vite integration. Strict gates and Codecov project/patch targets remain unchanged.
• 61/61 targeted mutants killed, including five worker invariants; 60,000 seeded property cases and forced-GC retention checks pass.
• Independent real consumers pass locally on Node 24.1.0, Bun 1.4.3, and Vite 8.3.4 / Playwright 1.64.0 in Chromium 156.0.8078.4, Firefox 157.0 and WebKit 27.2, each in development and production. Every browser run ends with zero live worker handles.
• All 12 CI jobs pass at signed head c444c5f against dev, including Deno 2.6.0/2.9.6, Node 22.12.0/24.1.0, Bun 1.4.3 and all six browser/build combinations. Downloaded CI artifacts confirm 2834/2834 lines, 1095/1095 branches and 253/253 functions, plus 61/61 mutants and zero remaining browser worker handles. Deno 2.9.6 on Linux reports two additional, fully covered branches at workers/entry.ts:4 compared with local Deno 2.9.7 on macOS; the source and strict gates are unchanged.
• Repository suites retain unit/integration filenames, BDD registrations and alias imports. Public factories, callable methods, options and error contracts have TSDoc; README examples are checked against the implementation.

Scope

This adds optional APIs to the released core, so the planned version remains 2.2.0. The replaced registry/pool API was introduced only in this unpublished PR. No WASM, automatic job retries, implicit buffer transfer, GC-based lifecycle guarantee, merge or publication is included. Custom transports must honor cancellation and actual termination; unconfirmed cleanup is reported. Module initialization runs in both caller and workers. Workers do not provide a security sandbox, and no workload-independent speedup is claimed. WebKit conformance is not a Safari-release certification.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.96%. Comparing base (af8220a) to head (c444c5f).

❌ Your project status has failed because the head coverage (99.96%) is below the target coverage (100.00%). You can increase the head coverage or adjust the target coverage.

Additional details and impacted files

@@             Coverage Diff             @@
##               dev      #40      +/-   ##
===========================================
- Coverage   100.00%   99.96%   -0.04%     
===========================================
  Files           17       39      +22     
  Lines         1998     3002    +1004     
  Branches       247      550     +303     
===========================================
+ Hits          1998     3001    +1003     
- Misses           0        1       +1     

Files with missing lines	Coverage Δ
src/context/runtime.ts	100.00% <100.00%> (ø)
src/error/index.ts	100.00% <100.00%> (ø)
src/error/meta.ts	100.00% <100.00%> (ø)
src/error/serialize.ts	100.00% <100.00%> (ø)
src/index.ts	100.00% <100.00%> (ø)
src/pipe/error.ts	100.00% <100.00%> (ø)
src/pipe/factory.ts	100.00% <100.00%> (ø)
src/pipe/index.ts	100.00% <100.00%> (ø)
src/pipe/pipe.ts	100.00% <100.00%> (ø)
src/plugin/error.ts	100.00% <100.00%> (ø)
... and 28 more

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.