Skip to content

Update enterpilot/gomodel Docker tag to v0.1.106 - #7933

Open
renovate[bot] wants to merge 1 commit into
localAppsfrom
renovate/enterpilot-gomodel-0.x
Open

renovate[bot] wants to merge 1 commit into
localAppsfrom
renovate/enterpilot-gomodel-0.x

Conversation

@renovate

@renovate renovate Bot commented Oct 22, 2026

Copy link
Copy Markdown

This PR contains the following updates:

Package Update Change
enterpilot/gomodel patch 0.1.96 → 0.1.106

Warning

Some dependencies could not be looked up. Check the Dependency Dashboard for more information.


Release Notes

ENTERPILOT/GoModel (enterpilot/gomodel)

v0.1.106

Compare Source

Changelog

Features
  • 99576a6 feat(decisions): add OpenAI Decisions API with Jev/Kev translation (#​1219)
  • 11e1771 feat(decisions): answer with OpenRouter decision models, warn on skipped response guardrails, add release scenarios (#​1227)
  • eedf88e feat(decisions): read images on hosted Jev and route images past text-only models (#​1220)
  • a53ca18 feat(systemone): accept multipart image requests (#​1221)
Bug Fixes
  • cb8aa00 fix(passthrough): require per-key passthrough access and provider grants for model-less calls (#​1218)
  • 51f0698 fix(server): implement ValidateProviderAccess on decisions test authorizer (#​1230)
  • 371e36b fix(server): remove duplicate ValidateProviderAccess on decisions test authorizer (#​1236)

Full Changelog: ENTERPILOT/GoModel@v0.1.105...v0.1.106

v0.1.105

Compare Source

Changelog

Bug Fixes
  • f9c4e22 fix(anthropic): replay server-side tool search on translated models (#​1214)

Full Changelog: ENTERPILOT/GoModel@v0.1.104...v0.1.105

v0.1.104

Compare Source

Changelog

Features

Full Changelog: ENTERPILOT/GoModel@v0.1.103...v0.1.104

v0.1.103

Compare Source

Changelog

Features
  • 1e4ecfb feat(ext): let extensions wrap every log handler core builds (#​1191)
Bug Fixes
  • a6b72f0 fix(storage): use ADD COLUMN IF NOT EXISTS on PostgreSQL migrations (#​1189)
  • 83a2c1c fix(usage): price Anthropic 1h cache writes and map streamed thinking tokens (#​1188)
Other Changes
  • 3ace958 tests(e2e): fixes before the previous release

Full Changelog: ENTERPILOT/GoModel@v0.1.102...v0.1.103

v0.1.102

Compare Source

Changelog

Features
Bug Fixes
  • d14e91f fix(dashboard): center the endpoint filter and whiten usage filter selects (#​1181)
  • eceb306 fix(providers): honor allowed_tools on Responses for chat-bridged providers (#​1156)
  • 380e03a fix(responses): drop tool_choice when no tools remain for chat translation (#​1175)

Full Changelog: ENTERPILOT/GoModel@v0.1.101...v0.1.102

v0.1.101

Compare Source

Changelog

Features
  • d99cdd3 feat(admin): accept secret references in dashboard-managed entities (#​1127)
  • eab7eb4 feat(config): rotate referenced secrets without restart (#​1126)
  • 832a09e feat(presidio): restore reasoning, per-tool and per-role restore, placeholder variants (#​1157)
  • d09be00 feat(server): turn LiteLLM tags into request labels (#​1152)
Bug Fixes
  • 25aa7c3 fix(admin): require the master key to repoint entities that hold secret references (#​1169)
  • 4a78059 fix(admin): require the master key to save secret references (#​1166)
  • 693add1 fix(app): keep the secret rotation watcher responsive when a resolver hangs (#​1148)
  • b8b6f01 fix(app): reject a reload that would drop a serving entity whose secret references fail (#​1168)
  • 548f347 fix(config): bound secret resolution so a hanging reference cannot wedge saves, rotation, or shutdown (#​1167)
  • bf86fe8 fix(config): reject empty secret files instead of resolving to an empty value (#​1162)
  • 9672d3e fix(dashboard): keep dialogs and model pickers usable on mobile Safari (#​1158)
  • 0031962 fix(modeldata): advertise capabilities from catalog input modalities (#​1153)
  • 70fc5db fix(presidio): keep restored values out of replayed reasoning and isolate placeholder slots (#​1164)
  • bddc9b8 fix(presidio): keep signed thinking unedited on streamed Messages (#​1163)
  • 4212dad fix(server): cap LiteLLM tags and allow turning them off (#​1161)

Full Changelog: ENTERPILOT/GoModel@v0.1.100...v0.1.101

v0.1.100

Compare Source

Changelog

Features
  • 621bdc8 feat(authkeys): import LiteLLM virtual keys by hash (#​1130)
  • 1e891f9 feat(cli): add gomodel migrate litellm config converter (#​1109)
  • 2083c6f feat(cli): import LiteLLM keys, teams, and budgets from its database (#​1132)
  • 439c868 feat(config): add secret references with env and file schemes (#​1123)
  • c45c370 feat(ext): admin-guarded extension routes and health checks (#​1125)
  • 04dafb8 feat(mcp): add virtual MCP servers (#​1120)
  • d26539f feat(openai): count Messages tokens exactly with OpenAI's input_tokens endpoint (#​1117)
  • f8c8b77 feat(storage): encrypt dashboard-managed secrets at rest (#​1124)
Bug Fixes
  • 2a64952 fix(chatgpt): forward prompt_cache_key and session-id to the Codex backend (#​1145)
  • 95a5067 fix(chatgpt): keep streamed output items in non-streaming responses (#​1147)
  • a63d4b1 fix(chatgpt): send prompt_cache_key as session-id when no session is detected (#​1146)
  • d22f9ec fix(models): retrieve models by any ID inference accepts and page the Anthropic list (#​1118)
  • a07fbbb fix(virtualmodels): skip disabled targets instead of failing the request (#​1137)

Full Changelog: ENTERPILOT/GoModel@v0.1.99...v0.1.100

v0.1.99

Compare Source

Changelog

Features
  • bc02d02 feat(dashboard): show MCP connect panel and tool discovery mode (#​1111)
  • ebaad34 feat(mcp): add opt-in tool search discovery (#​1102)
  • cd128a9 feat(openai): serve chat requests through Responses when Chat Completions cannot (#​1112)
  • d6c6f96 feat(providers/kimicode): add native Responses API support (#​1083)
  • 0866628 feat(telemetry): export GenAI usage and opt-in content on provider spans (#​1115)
  • 689e787 feat: add Eden AI provider integration (#​934)
Bug Fixes
  • 05c0f16 fix(anthropicapi): send text documents to OpenAI as text, report refusals and upstream stream errors (#​1116)
  • 7566ec8 fix(openai): adapt chat params for GPT-6 and tool calls on GPT-5.6+ (#​1108)
  • 9113022 fix(openai): adapt stop, top_p and reasoning effort for reasoning models (#​1114)
  • eef1614 fix(registry): keep serving during model discovery outages (#​1113)
  • 15054f4 fix(usage): report OpenAI cache tokens in Anthropic usage and price cache writes (#​1110)

Full Changelog: ENTERPILOT/GoModel@v0.1.98...v0.1.99

v0.1.98

Compare Source

Changelog

Features
  • 5d9af07 feat(jev): add native /v1/systemone endpoint (#​1097)
  • 667453c feat(jev): cache, fail over, pin versions, and serve Kev routes on /v1/systemone (#​1098)
  • 1d1c3f0 feat(mcp): per-server tool and user-path exclusions (#​1094)
Bug Fixes
  • e9d274e fix(auth): keep header user path on MCP, realtime, and audio endpoints (#​1096)
  • 12861be fix(jev): record System One token totals and prices, allow pinned versions in virtual models (#​1099)
  • 2f1a0e5 fix(providers): honor allowed_tools, strict and developer role on Gemini and Anthropic (#​1091)
  • db5e53c fix(providers): treat missing /models endpoint as healthy when models are configured (#​1092)

Full Changelog: ENTERPILOT/GoModel@v0.1.97...v0.1.98

v0.1.97

Compare Source

Changelog

Features
Bug Fixes
  • b8cb81c fix(dashboard): drop the details accordion from virtual model rows (#​1079)
  • 6140859 fix(tests): skip .worktrees in testconventions scan (#​1087)
  • be89ba0 fix(workflows): return compile errors from GetView instead of 502 (#​1088)

Full Changelog: ENTERPILOT/GoModel@v0.1.96...v0.1.97


Configuration

📅 Schedule: (UTC)

  • Branch creation
    • At any time (no schedule defined)
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Never, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.


  • If you want to rebase/retry this PR, check this box

This PR was generated by Mend Renovate. View the repository job log.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants