Changelog

Features

• 7e481fe feat(meta): add Meta Model API provider for Muse Spark models (#520)
• 20977e1 feat(providers): rotate requests across multiple API keys per provider (#516)
• 44f0b70 feat(xai): support grok-4.5 reasoning effort mapping (#519)

Other Changes

• 239cf11 Add PostHog analytics integration to docs

Changelog

Huge thank you to @weselben for implementing Kimi Code provider support!

Features

• 47cc070 feat(audit): show rewrite token savings on the Rewritten tab (#507)
• a94d01e feat(dashboard): request status code and provider latency charts (#501)
• 618cf40 feat(dashboard): stream in-flight response bodies into live logs and interactions (#495)
• 4bd78f9 feat(provider): add Kimi Code provider (#508)
• 2c3ab68 feat(realtime): add OpenAI-compatible WebRTC voice endpoints (#500)
• e0eb0ee feat(usage): self-service usage and limit status endpoint (#498)
• f157d01 feat(virtualmodels): validate target providers at startup (#494)

Bug Fixes

• 8e53f67 fix(config): reject unknown config keys and log provider origins (#511)
• 23e3113 fix(dashboard): hide Rate Limits nav item when the feature is disabled (#512)
• 874aa02 fix(lifecycle): close leaked goroutines, sessions, and unbounded error reads (#497)
• 383248d fix(ratelimit): honest 429 when every virtual-model target is saturated (#492)
• ebd8320 fix(realtime): resolve virtual-model aliases on realtime and audio endpoints (#514)

Performance

• 92db958 perf(memory): stop pinning request-sized buffers for the stream lifetime (#489)

Other Changes

• 37dcd7f tests(e2e): added fireworks tests and a few more improvements

Changelog

Features

• 49f895e feat(fireworks): add Fireworks AI provider (#475)
• 6b1728b feat(ratelimit): scoped request, token, and concurrency limits (user paths, providers, models) (#482)
• 1085323 feat(usage): track prompt tokens and cost saved by request rewriters (#481)
• 9e21e80 feat: public extension API and embeddable bootstrap (#479)

Bug Fixes

• eb08268 fix(anthropicapi): carry top_p/user in typed fields + architecture-review quick wins (#472)
• 54b3429 fix(config): honor the http: YAML block for upstream client timeouts (#483)
• 5b885dd fix(docs): make Discord nav link blue (#471)
• 832c0ed fix(docs): set Mintlify theme colors to blue (#470)
• a069a2c fix(memory): bound live-log retention and persist responses/conversations (#488)
• 72246d5 fix(resilience): circuit breaker correctness, stale provider inventory, fast health rechecks (#480)

Changelog

Features

• 3c817e8 feat(authkeys): labels on managed API keys applied to every request (#467)
• 1d7cfe5 feat(dashboard): label analytics and page-level filters on usage page (#461)
• 4c8e0d7 feat: added labelling system (#454)

Bug Fixes

• 4a951ae fix(gateway): skip failover sweep when request context is done (#453)
• 3573295 fix: redact provider api keys, bound cache writers, harden pgvector init (#452)

Performance

• 7a052e8 perf: cut hot-path allocations across streaming, live logs, and request adaptation (#456)

Other Changes

• 87d78d2 tests(e2e): added new labelling related tests to the set

Changelog

Features

• 92d4e88 feat(dashboard): refine activity calendar gradation, colors, alignment + chart day markers (#443)
• abc9129 feat: add manual failover management (#444)

Other Changes

• 33e4d75 tests: better test coverage for failover (#450)

Changelog

Bug Fixes

• 236c2f0 fix(usage): apply cached-token discounts to all OpenAI-compatible providers (#439)

Changelog

The overview page has been changed. It surfaces prompt caching stats and live preview:

Also, load balancing has been added to virtual models feature.

More details:

Features

• f291083 feat(dashboard): add token cache meter and overview refinements (#428)
• 34f13d6 feat(dashboard): live token throughput chart, prompt cache gauge, and overview refinements (#434)
• 9502487 feat(virtualmodels): load-balanced virtual models (round-robin + cost) with IaC config (#433)

Bug Fixes

• c8b6a80 fix(virtualmodels): don't gate IaC startup on catalog availability (#436)

Changelog

In this release, we unified aliases with model overrides because, technically, they are the same thing. The biggest changes from the user perspective are:

1. The admin API for aliases and model overrides changed to virtual models.
2. Aliases are now called virtual models.

Thank you, @vfeitoza, for your PR, which inspired me to make this simplification. The reported issue is now fixed by design: #387

The detailed list of changes below:

Breaking Changes

• 6b29627 feat(virtualmodels)!: unify engine, admin API, and dashboard UI (#423)

Features

• b113cab feat(admin): refine dashboard model and overview UI (#426)
• df9d185 feat(virtualmodels): unify aliases and access overrides into virtual models (#416)

Bug Fixes

• 787f878 fix(dashboard): scale usage calendar intensity by daily volume (#425)

Changelog

Bug Fixes

• 9025183 fix(streaming): normalize non-SSE upstream chat completions (#412)

Performance

• 2677c1f perf: faster model resolution, JSON decoding, and request snapshotting (#413)

Changelog

Features

• faf65b1 feat(dashboard): support fully offline use (#408)