What ENTERPILOT/GoModel shipped
Written by FoxPlug from the repository's own releases; not affiliated with Enterpilot. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/ENTERPILOT/GoModel; Enterpilot did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- V0.2.3 released with a breaking change: a configured provider model list now exposes only those models by default, and CONFIGURED_PROVIDER_MODELS_MODE=fallback restores the old behavior. Release
- Operators can sign in to OAuth-protected upstream MCP servers such as Notion, Linear and Atlassian from the dashboard MCP Servers page, with servers answering 401 marked sign-in required. Pull request #1278
- MCP sessions issued by one replica now work on every replica sharing the same PostgreSQL or MongoDB database, so clients no longer need sticky sessions. Pull request #1340
- Virtual models and model policies accept optional params and
default_paramsmaps, so one self-hosted model can be exposed under several aliases with their own sampling and thinking settings. Pull request #1353 - Rate limits for requests, tokens and concurrency now hold across all replicas sharing a database instead of each replica admitting the full limit. Pull request #1339
- An opt-in cache setting gives each effective user path its own entries in both the exact and the semantic response cache. Pull request #1352
- The dashboard live log view now shows traffic from every replica rather than only the replica the browser is connected to. Pull request #1341
- The Playground gets an Embeddings mode that posts to /v1/embeddings through the gateway, so requests appear in Audit Logs and Usage. Pull request #1386
- Providers gain a per-provider billable flag, default true, so budgets and billing skip usage on providers where the caller does not owe the operator, such as bring-your-own-key. Pull request #1349
- OpenRouter image generations now record their cost and input and output token counts instead of being saved as $0 with zero tokens. Pull request #1383
Why it matters
This week brings v0.2.3, which changes the default for configured provider model lists, along with sign-in for OAuth-protected MCP servers and MCP sessions that work across replicas. Several fixes touch everyday use: image costs on OpenRouter now show up in usage, live logs cover every replica, and rate limits are shared across replicas on one database.
Changelog entry
- V0.2.3: breaking change, configured provider model lists now expose only those models by default; CONFIGURED_PROVIDER_MODELS_MODE=fallback restores the old behavior. Release
- The default configured provider models mode is now allowlist, so a configured model list is the provider's whole inventory unless changed. Pull request #1384
- Added a per-provider
models_modeoverride so one provider's model list can be limited without switching the global mode. Pull request #1376 - Allowlisted models now take provider metadata such as context window, max output, pricing and model type from the provider's /models endpoint. Pull request #1385
- Dashboard MCP Servers page can sign in to OAuth upstream MCP servers, following the MCP authorization spec. Pull request #1278
- Downstream MCP sessions resumed on any replica sharing the same PostgreSQL or MongoDB database. Pull request #1340
- Request, token and concurrency rate limits now hold across replicas that share a database. Pull request #1339
- Virtual models and model policies accept params and
default_paramsmaps to set request parameters per alias. Pull request #1353 - Opt-in response cache partitioning by user path, for both the exact and the semantic cache. Pull request #1352
- Live log view shows traffic from every replica, not only the connected one. Pull request #1341
- Playground adds an Embeddings mode that posts to /v1/embeddings through the gateway. Pull request #1386
- OpenRouter image generations now record cost and input and output token counts. Pull request #1383
V0.2.3 is out. Configured provider model lists now expose only those models by default; set CONFIGURED_PROVIDER_MODELS_MODE=fallback to keep the old behavior. Also this week: OAuth sign-in for MCP servers from the dashboard, MCP sessions that work on any replica, and OpenRouter i
This week we released v0.2.3 for GoModel. The release includes a breaking change: a configured provider model list now exposes only those models by default, and setting CONFIGURED_PROVIDER_MODELS_MODE=fallback restores the previous behavior. Alongside it, operators can sign in to OAuth-protected MCP servers from the dashboard's MCP Servers page, MCP sessions issued by one replica now work on every replica sharing a database, and rate limits hold across replicas. On the usage side, OpenRouter image generations now record cost and token counts instead of showing zero.
Week of October 12, 2026
What shipped
- Usage can now be broken down and filtered by endpoint, with a new admin endpoint returning requests, tokens, costs and cache fields per endpoint. Pull request #1178
- Releases now attach a prebuilt dashboard tarball listed in the checksums file, so package maintainers and downstream builds can unpack it and build without a separate dashboard build step. Pull request #1180
- Extensions can now wrap every log handler that core builds, so log redaction keeps working across reloads rather than only on the default logger. Pull request #1191
- PostgreSQL column migrations now use ADD COLUMN IF NOT EXISTS, so restarts no longer write an error per already-applied column to the server log. Pull request #1189
- The two inlang plugins are now installed through npm and covered by the lockfile, so dashboard builds work offline in sandboxed environments. Pull request #1179
- Version 0.1.103 was released with the extension log handler wrapping, the PostgreSQL migration fix and the usage pricing fix. Release
- The endpoint filter on the Usage page is now vertically centered and the usage filter selects use the surface background. Pull request #1181
- Lint now restores Go caches through the shared cache action, and the voice latency guard was tightened. Pull request #1190
- The Discord release announcement no longer posts raw commit SHAs, links pull request references and truncates long bodies at the last complete change. Pull request #1197
Why it matters
This week adds endpoint-level visibility into usage, so you can see what chat, audio or embeddings cost. It also makes releases easier to consume for anyone building from source, and fixes pricing, logging and migration behavior that affected day-to-day operation.
Changelog entry
- Break usage down and filter it by endpoint, with a new admin endpoint for per-endpoint requests, tokens, costs and cache fields. Pull request #1178
- Publish the prebuilt dashboard tarball with each release, listed in checksums.txt. Pull request #1180
- Let extensions wrap every log handler core builds. Pull request #1191
- Use ADD COLUMN IF NOT EXISTS on PostgreSQL migrations to stop error lines on restart. Pull request #1189
- Install inlang plugins through npm for offline dashboard builds. Pull request #1179
- Center the endpoint filter and apply surface background to usage filter selects. Pull request #1181
- Refresh the Go cache for lint through the shared cache action and tighten the voice latency guard. Pull request #1190
- Strip SHAs, link pull request references and line-truncate the Discord release announcement. Pull request #1197
- Release v0.1.103. Release
- Release v0.1.102. Release
Week of September 14, 2026
What shipped
- V0.1.95 shipped with audio.cpp provider support, master key authentication disable option, and configurable trusted proxy resolution for client addresses. Release
- Audio.cpp provider added for local
audiocpp_server, supporting /v1/audio/speech and /v1/audio/transcriptions routes. Pull request #1058 - Master key authentication can now be disabled via MASTER_KEY_DISABLED setting, forcing deployments to use managed gateway API keys exclusively. Pull request #1067
- Trusted proxy CIDR configuration replaced the previous LOGGING_TRUSTED_PROXY_CIDRS setting with three new server.* configuration options. Pull request #1061
- Audio speech and transcription responses now stream instead of buffering, reducing time-to-first-byte. Pull request #1057
- Streaming requests now fail over when the provider sends nothing or opens with an in-band error, instead of reaching clients as silent EOF. Pull request #1015
- HTTP_STREAM_IDLE_TIMEOUT setting added to cap how long a stalled provider stream can hold connections, separate from HTTP_TIMEOUT. Pull request #1018
- Truncated provider streams now end with an explicit error event instead of silent EOF. Pull request #1019
- Provider attempt rows in audit logs now insert as a single multi-row statement per chunk instead of individual queries. Pull request #1036
- Sticky session pins and realtime call registry now use expiry indexes instead of full-map sweeps on every insert. Pull request #1027
Why it matters
V0.1.95 adds audio.cpp provider support and improves operational control over authentication and client address logging. Streaming reliability improvements and performance optimizations reduce latency and resource usage under failover-heavy workloads.
Changelog entry
- V0.1.95 Release
- Feat(audiocpp): add the audio.cpp provider for local
audiocpp_serverPull request #1058 - Feat(server): add MASTER_KEY_DISABLED option to disable master key authentication Pull request #1067
- Feat(server): replace LOGGING_TRUSTED_PROXY_CIDRS with three server.* trusted proxy settings Pull request #1061
- Fix(audio): stream speech and transcription bodies instead of buffering Pull request #1057
- Fix(llmclient): fail over when a stream opens empty or with an in-band error Pull request #1015
- Fix(llmclient): add HTTP_STREAM_IDLE_TIMEOUT to cap stalled provider streams Pull request #1018
- Fix(streaming): end truncated streams with an explicit error event Pull request #1017
- Perf(auditlog): insert provider attempts in one statement per chunk Pull request #1036
- Perf(routing): evict from an expiry index instead of scanning Pull request #1027
- Refactor(providers): inject the test transport through ProviderOptions Pull request #1039
V0.1.95: audio.cpp provider, master key disable option, trusted proxy config. Streaming responses no longer buffer. Stalled streams fail over instead of hanging. Audit inserts batch provider attempts.
V0.1.95 ships with three operational improvements: audio.cpp provider support, configurable master key authentication, and trusted proxy address resolution. Streaming reliability is stronger—idle streams fail over automatically and responses no longer buffer. Audit logging batches provider attempts, reducing database round trips on failover-heavy batches.
More from ENTERPILOT/GoModel on GitHub
Merged pull requests
- fix: close the round-15 pre-release findings (#1391, merged November 10, 2026 by SantiagoDePolonia) (source)
- docs(guides): enable OpenCode reasoning effort variants for GoModel models (#1398, merged November 10, 2026 by mansur-akhmadi) (source)
- fix: close the 0.2.3 pre-release findings (#1389, merged November 8, 2026 by SantiagoDePolonia) (source)
- fix(providers): throttle on-demand model list refreshes for unknown models (#1362, merged November 8, 2026 by SantiagoDePolonia) (source)
- perf(cache): hash the exact-cache key once per request and bound stream capture (#1363, merged November 8, 2026 by SantiagoDePolonia) (source)
- feat(models): take provider metadata for allowlist models from /models (#1385, merged November 8, 2026 by SantiagoDePolonia) (source)
- fix(cluster): bound background database polls with per-tick deadlines (#1372, merged November 8, 2026 by SantiagoDePolonia) (source)
- perf(streaming): scan SSE event boundaries in one linear pass (#1361, merged November 8, 2026 by SantiagoDePolonia) (source)
- feat(auditlog): add LOGGING_LOG_EMBEDDING_BODIES to keep embedding bodies out of the audit log (#1387, merged November 8, 2026 by urtho) (source)