Skip to content

v1.112.0-dev.1

Pre-release
Pre-release

Choose a tag to compare

@litellm-release litellm-release released this 20 Nov 00:03
· 31 commits to main since this release
07a338c

Verify Docker Image Signature

cosign verify \
  --certificate-identity https://github.com/BerriAI/project-releaser/.github/workflows/release.yml@refs/heads/main \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com \
  ghcr.io/berriai/litellm:v1.112.0-dev.1

What's Changed

  • test(database): expect no native pool in proxy lifespan cleanup by @devin-ai-integration[bot] in #51911
  • fix(ui): clarify customer usage periods and preserve date ranges by @moe-berri in #51899
  • feat(ui): reuse usage spend design for customer teams and members by @tin-berri in #51724
  • feat(customers): move existing members between teams by @moe-berri in #51913
  • feat(customers): search customer members by ID or alias in the Members tab by @kenneth-berri in #51896
  • test(e2e): drop two batch waits that never finish in time by @yuneng-berri in #51882
  • test(integration): record a stack when an owned proxy crashes by @devin-ai-integration[bot] in #51922
  • refactor(s3-log): move the rust-backed s3 logger from s3_v3 to s3_v2 by @narmstrong-berri in #51893
  • perf(host-python): settle ready async values inline instead of spawning on tokio by @narmstrong-berri in #51906
  • perf(streaming): extract usage once per chunk when assembling stream usage by @narmstrong-berri in #51903
  • perf(caching): format dual cache debug logs lazily by @narmstrong-berri in #51902
  • perf(auth): read token model fields from the class instead of deprecated fields by @narmstrong-berri in #51901
  • refactor(s3-log): move python-compatible helpers into python-compat by @narmstrong-berri in #51895
  • fix(proxy): join default teams on the team member budget when the team has one by @kerry-berri in #51728
  • fix(proxy): let an exhausted key keep listing models and counting tokens by @kerry-berri in #51675
  • fix(proxy): run the /fal_ai passthrough price gate on the path sent upstream by @kerry-berri in #51671
  • fix(fal_ai): send versioned seedream and flux-pro image models to the endpoint they name by @kerry-berri in #51668
  • fix(replicate): answer malformed video seconds with a 400 and drop non-decimal size digits instead of a 500 by @kerry-berri in #51670
  • fix(perplexity): bill cached prompt tokens at the cache read rate when the response has no cost by @kerry-berri in #51669
  • fix(vertex_ai): send extra_body fields as predict parameters for BGE embedding models by @kerry-berri in #51662
  • fix(bedrock): put extra_body fields inside modelInput for async-invoke embeddings by @kerry-berri in #51676
  • fix(fal_ai): bill a video request without a resolution at the row's default rate by @kerry-berri in #51665
  • fix(router): stop the rust router keeping every finished request in memory until a full gc by @mateo-berri in #51888
  • fix(customers): keep uncapped customer spend across scheduled budget resets by @moe-berri in #51918
  • fix(budgets): match window starts to stored millisecond precision by @moe-berri in #51921
  • fix(ui): default customer usage range to month to date by @moe-berri in #51923
  • fix(cost): keep a deployment's per-second video price over the catalog row's flat price by @kerry-berri in #51688
  • feat(dynamic_rate_limiter): customer organizations and keys borrow idle model capacity and split class shares by @kerry-berri in #51705
  • fix(count_tokens): return 400 for malformed Anthropic server-tool blocks by @kerry-berri in #51957
  • fix(admin-mcp): restore the budget route operation ids the connector pins by @kerry-berri in #51959
  • fix(proxy): hold the logging object weakly in the detached stream failure hook by @kaizen-38 in #51318
  • fix(responses): retry a stream that breaks before output in its own model group by @kerry-berri in #51953
  • perf(customers): index member team, budget and override lookups by @moe-berri in #51961
  • feat(customers): sort organizations, teams and members by budget limit by @moe-berri in #51924
  • chore(lint): flag conditionals with more than 3 boolean clauses via ruff PLR0916 by @mateo-berri in #51964
  • refactor(vertex_ai): remove unreachable VertexAIFilesHandler.afile_content by @mateo-berri in #51966
  • fix(router): free a failed rust router call's request without waiting for a full gc by @mateo-berri in #51963
  • perf(auth): reuse validated models on UserApiKeyCache memory hits by @narmstrong-berri in #51912
  • fix(proxy): reset members given default team member models on the team's budget window by @kerry-berri in #51960
  • fix(customers): cascade delete large organizations within the transaction timeout by @moe-berri in #51969
  • feat(budgets): replace the python budget hold and settle engine with a rust engine admitted at auth by @narmstrong-berri in #51535
  • fix(dynamic_rate_limiter): charge tokens used while the model was idle against a team's class share by @kerry-berri in #51952
  • feat(messages): serve bedrock invoke messages on rust only by @narmstrong-berri in #51330
  • fix(proxy): stop team model writes from pulling in proxy-level settings by @shrey-berri in #51874
  • fix(proxy)!: match the sign-in email fallback to the right user by @shrey-berri in #51897
  • fix(messages): clamp bedrock effort before the drop_params effort gate by @narmstrong-berri in #51972
  • refactor(proxy): keep the detached stream failure hook state on the logging object by @narmstrong-berri in #51962
  • feat(fnmatch): add a crate for Python's fnmatchcase glob dialect by @yujonglee-berri in #51530
  • refactor(auth): decide route rules in Rust by @yujonglee-berri in #51531
  • refactor(auth): plan the authentication path in Rust by @yujonglee-berri in #51605
  • fix(decisions): answer 400 naming extra_headers when it is not an object by @paveln83 in #50581
  • fix(proxy): stop returning stack traces in health check errors by @ojensen-berri in #52005
  • chore(cost-map): add fal_ai seedance 2.5 flash rows, copilot deprecation dates, inkling 2 small rows, pixverse clip prices and groq/nebius fixes by @devin-ai-integration[bot] in #51929
  • chore(cost-map): add fal_ai/bria/rmbg-3 by @devin-ai-integration[bot] in #52041
  • fix(proxy): prevent silent failures to save custom model prices by @AaronHowell in #48746
  • chore(cost-map): add openai omni-moderation-csam-latest by @devin-ai-integration[bot] in #52047
  • feat(control_plane): service skeleton, settings, license gate, JWT auth and scope RBAC by @yassin-berriai in #49654
  • feat(control_plane): dataplane registry, credential strategies and client by @yassin-berriai in #49655
  • feat(control_plane): fan-out primitive and global usage and fleet aggregates by @yassin-berriai in #49656
  • feat(control_plane): authenticated management passthrough and audit log by @yassin-berriai in #49657
  • feat(control_plane): --control-plane flag serves the control plane API from the existing image by @yassin-berriai in #49658
  • feat(control_plane): otel server and dataplane client spans with w3c propagation by @yassin-berriai in #49659
  • feat(control_plane): ocsf audit events, sink protocol, queued delivery and hash chain by @yassin-berriai in #49660
  • feat(control_plane): s3 object lock audit sink by @yassin-berriai in #49661
  • feat(control_plane): splunk hec and microsoft sentinel audit sinks by @yassin-berriai in #49662
  • feat(control-plane): runtime dataplane and role writes via redis stream and postgres by @yassin-berriai in #51762
  • refactor(rust): read config values and YAML with Python semantics by @yujonglee-berri in #52023
  • feat(config-load): resolve os.environ/ references for the gateway by @yujonglee-berri in #52025
  • refactor(rust): drop os.environ/ stripping where config-load already resolved it by @yujonglee-berri in #52026
  • refactor: drop fresh suppressions in rate limiter and rust streaming, type perplexity cache read by @mateo-berri in #51996
  • test(sso): reuse the provider identity across dashboard and CLI sign-in by @devin-ai-integration[bot] in #52062
  • fix(images): pass the caller's OpenAI image params to custom provider handlers by @kerry-berri in #52051
  • fix(embeddings): let extra_body replace Bedrock embedding defaults and combine nested objects by @kenneth-berri in #51710
  • test: remove dead skip-marked tests and rewrite the Redis buffer drain test by @yuneng-berri in #52063
  • fix(gemini): send reasoning_effort xhigh as thinkingLevel xhigh for gemini-4-hydrogen by @kerry-berri in #52060
  • fix(logging): retain streaming success logging tasks until they finish by @joshua-berri in #52067
  • perf(token_counter): memoize token counts of long texts per tokenizer by @narmstrong-berri in #51904
  • fix(otel v2): stop legacy langfuse loggers sending to refused key/team hosts by @shivamrawat1 in #51344
  • fix(router): preserve request tags with pre-call guardrails by @shivamrawat1 in #50488
  • test(database): expect the separate daily spend index build by @devin-ai-integration[bot] in #51504
  • test(budgets): give the customer reset fixtures an explicit cap by @devin-ai-integration[bot] in #52058
  • test(spend): expect model hub defaults on pass-through endpoints by @devin-ai-integration[bot] in #51505
  • test(routing): expect zero admission tokens in priority headers by @devin-ai-integration[bot] in #52065
  • fix(gemini): return 400 when an image edit request has no image by @kenneth-berri in #52087
  • test(auto_router): verify total_spend in session integration responses by @joshua-berri in #51283
  • perf(openai): decode streamed chat chunks with pydantic's compiled validator by @narmstrong-berri in #51905
  • docs(custom-logger-v2): add a user guide for writing v2 callbacks by @narmstrong-berri in #52076
  • fix(embeddings): apply additional_drop_params to provider-specific params by @kenneth-berri in #52098
  • feat(fal_ai): add 8s clip prices for Pixverse effects, transition and extend by @kenneth-berri in #52107
  • fix(responses): stream tool-call arguments that only arrive in done events by @kenneth-berri in #52069
  • fix(messages): close high and medium rust messages parity bugs by @narmstrong-berri in #52042
  • test: update four stale integration and UI tests for intended behavior changes by @yuneng-berri in #52108
  • feat(bedrock): add us. and global. cost map rows for Daybreak Blue and Daybreak Red by @jesus-berri in #52122
  • test(spend): wait for the scheduled reset to commit before reading the reset row by @devin-ai-integration[bot] in #52128
  • fix(proxy): serve a burst of first requests on new customer org aliases instead of timing out Prisma transactions by @kerry-berri in #52080
  • fix(dynamic_rate_limiter): keep counting and reporting priority requests after a priority pool's tokens run out by @kerry-berri in #52081
  • fix(proxy): check pre-call fallback access with the saved model_group_alias, never the one the client sent by @kerry-berri in #52077
  • fix(proxy): resolve admin MCP authorization without listing every key and team by @devin-ai-integration[bot] in #52096
  • fix(ui): generate dashboard IDs without crypto.randomUUID so plain HTTP works by @moe-berri in #52129
  • feat(cost): add automatic ECB reference-rate lookup by @joshua-berri in #51934
  • fix(proxy): preserve terminal evidence in disconnect billing by @joshua-berri in #52088
  • feat(alerts): display prompt cache advisories in existing alerts by @joshua-berri in #51943
  • fix(gemini): return n images for image generations and edits when n > 1 by @kenneth-berri in #52103
  • fix(proxy): preserve team aliases in realtime websocket routing by @norkelbyte in #52124
  • chore(cost-map): add Azure US Government rows for gpt-5.6, gpt-4.1, gpt-4o and ada-002 by @kerry-berri in #52135
  • fix(proxy): rewrite batch JSONL alias rows to the deployment model on model-routed uploads by @mika-lt in #51843
  • feat(proxy): accept request token prices in a declared currency by @joshua-berri in #51971
  • fix(guardrails): log requests that a guardrail blocks on /apply_guardrail by @abesticode in #51014
  • fix(model_prices): rolling registry audit 2026-11-16, add dashscope qwen4-max, max reasoning effort for ChatGPT GPT-5.6 by @devin-ai-integration[bot] in #52097
  • feat(alerts): add email labels, bulk actions and threshold editing by @moe-berri in #52139
  • feat(alerts): report observed Anthropic cache misses as advisories by @joshua-berri in #51973
  • feat(focus): upgrade FOCUS export to the 1.5 Cost and Usage schema by @mateo-berri in #52105
  • fix(router)!: stop a client's metadata flag from turning off silent_model mirroring by @mika-lt in #52083
  • chore(deps): bump next from 16.4.1 to 16.4.5 in /litellm/ui/litellm-dashboard by @dependabot[bot] in #52102
  • feat(ui): show an optional currency estimate on the Usage spend card by @joshua-berri in #51979
  • docs(rust): add python porting rules to rust agents.md by @narmstrong-berri in #52154
  • fix(caching): log a cache hit with the deployment's vertex_location by @kenneth-berri in #52150
  • fix(bedrock): log the prompt, settings and stream flag of passthrough requests by @kenneth-berri in #52146
  • test(observability): allowlist the key Langfuse collector host in the OTel v2 console test by @devin-ai-integration[bot] in #52157
  • fix(baseten): sync Kimi K3.1 Pro input limit by @devin-ai-integration[bot] in #52158
  • chore(cost-map): add successor and deprecation source to the github_copilot rows retired on 2026-11-16 by @devin-ai-integration[bot] in #52160
  • feat(mcp): add Parallel to the MCP discovery registry by @yucheng-berri in #52141
  • fix(opencode_go): forward chat reasoning effort by @moe-berri in #52156
  • fix(azure_ai): send reasoning_effort to Foundry reasoning models by @zoveki in #51775
  • fix(bedrock): recognize the ca. cross-region inference profile prefix by @kenneth-berri in #52152
  • fix(mcp): let zero-budget users call MCP tools that cost nothing by @kerry-berri in #52130
  • fix(mcp): keep an MCP server's own Authorization key when the JWT signer is on by @jesus-berri in #51867
  • feat(ui): add bulk actions for virtual keys by @tin-berri in #52138
  • fix(migrations): bring customer hierarchy tables in line with schema.prisma by @yuneng-berri in #52175
  • fix(auth): apply current team limits in the pre-body rate limit check by @yuneng-berri in #52169
  • refactor: share Replicate HTTP client factories and declare PagerDuty type alias by @mateo-berri in #51594
  • feat(harness): add typed agent evaluations (#52143) by @ishaan-j-berri in #52174
  • fix(bedrock)!: split the OpenAI breakpoint and Bedrock cachePoint model flags by @joshua-berri in #52180
  • fix(ui): drop format:date from Aria Agent tool schemas by @moe-berri in #52189
  • fix(bedrock): send guarded_text parts in system and assistant messages as Converse guardContent by @zoveki in #51776
  • fix(exceptions): map upstream 403 to PermissionDeniedError for OpenAI-style providers by @ryan-crabbe-berri in #52179
  • fix(complexity_router): classify short asks in long conversations by @moe-berri in #52183
  • test(providers): expect batch alias rows rewritten to the deployment model on model-routed uploads by @devin-ai-integration[bot] in #52199
  • fix(gemini): parse tool call arguments with the shared helper by @stanley-zhou in #50020
  • test(security): run the owned security proxy in offline mode so the ECB rate lookup never leaves the rig by @devin-ai-integration[bot] in #52202
  • fix(spend-logs)!: match key alias filter exactly through the api_key index by @ryan-crabbe-berri in #47795
  • fix(bedrock): route bedrock// to that region on /v1/messages by @kerry-berri in #52136
  • fix(messages): load the anthropic beta map only when a filter has a beta to filter by @maciej-berri in #52188
  • fix(gemini): keep Google's reason on non-streamed content_filter choices by @ryan-crabbe-berri in #52208
  • fix(anthropic): close /v1/messages thinking blocks with a signature and keep bridged output by @paveln83 in #52116
  • fix(ui): use full page navigation for SSO login by @kerry-berri in #52213
  • refactor(auth): port key and JWT authentication in a Rust core by @yujonglee-berri in #51374
  • fix(cost-map): declare mid-conversation system support on Databricks Claude Opus 5.5 by @kerry-berri in #52214
  • fix(skills)!: preserve legacy Skills Hub capabilities by @tin-berri in #51977
  • feat(router): select conversation headers for session affinity by @joshua-berri in #52193
  • feat(shadow-eval): skip redundant reverse evaluations on opt-in by @joshua-berri in #52192
  • fix(gemini): stop replaying a client-mangled thought signature from a tool call id by @ryan-crabbe-berri in #52207
  • feat(proxy): support Cursor CLI subscriptions by @moe-berri in #52205
  • fix(router): honor fallback opt-out for 413 and 422 bad requests by @joshua-berri in #52195
  • fix(copilot): accept empty or whitespace stored key with key ref (#52014) by @vanshsingh3711-wq in #52032
  • fix(router): keep both errors when a streamed cross-group fallback also fails by @joshua-berri in #52217
  • fix(responses): write one failure row per Responses WebSocket turn the deployment refuses by @quietjules in #50731
  • refactor(ui): build the dashboard and chat shells on the shadcn Sidebar by @yuneng-berri in #51257
  • fix: keep LiteLLM-internal params out of provider request bodies by @fangkangmi in #46083
  • fix(gemini): keep thought signatures for proxy model names without gemini in them by @ryan-crabbe-berri in #52212
  • feat(proxy): opt-in prefix block hashes in spend logs for workload replay by @mateo-berri in #52120
  • fix(rust): read a dict subclass by its items when converting Python input by @mika-lt in #51812
  • feat(callbacks-parallel-request-limiter): port parallel request limiter v3 to a rust callback by @narmstrong-berri in #51537
  • test(cost): mark the cost tracking cases the reconciliation automation bills against the real provider by @kerry-berri in #52226
  • feat(cost-map): price Bedrock Mantle models under bedrock_mantle/ and bedrock/mantle/ by @devin-ai-integration[bot] in #52231
  • refactor: reuse the shared session id reader and drop fresh restating comments by @mateo-berri in #52250
  • feat(litellm): add Vispark provider with Vision Small/Medium/Large by @vansh-one in #41192
  • feat(tts)!: stream audio from the provider as it arrives on /v1/audio/speech by @kabooboo in #46222
  • chore(cost-map): rolling registry audit 2026-11-17, Databricks gemini-4-hydrogen, fal_ai iris-z and H3 Max, Bedrock EU embed-v4, Vertex multimodalembedding by @devin-ai-integration[bot] in #52162
  • fix(together_ai): declare reasoning_effort_levels on Kimi-K3.1-Pro so max is not sent as high by @jesus-berri in #52294
  • fix(anthropic, bedrock): split concatenated tool call arguments and answer every split tool use by @kenneth-berri in #52216
  • fix(router)!: let an unmatched content_policy_fallbacks list fall through like context_window_fallbacks by @6matt in #52132
  • fix(cost-map): add the 1080p rate to the base fal_ai seedance 2.5 rows by @devin-ai-integration[bot] in #52304
  • fix(interactions): retain async stream success logging task through GC and shutdown drain by @joshua-berri in #52300
  • test(providers): use a valid base64 Gemini signature fixture in replay tests by @devin-ai-integration[bot] in #52311
  • fix(cohere)!: map tool_choice, reasoning_effort and json_schema response_format on the v2 chat route by @paveln83 in #50834
  • feat(guardrails): scan an image on /apply_guardrail with a Bedrock guardrail by @abesticode in #51016
  • fix(gateway): map OpenAI chat params to provider names before the Rust core checks them by @tavren27 in #52211
  • feat(cost-map): add azure gpt-6.2-astra, gpt-6.2-sol and azure_ai/Microsoft-Decision-1.1 by @devin-ai-integration[bot] in #52293
  • feat(cost-map): add openai gpt-6.2-astra, gpt-6.2-sol and gpt-6.2-astra-law by @devin-ai-integration[bot] in #52316
  • fix(prompt_management): apply a prompt manager once per request on acompletion and SDK fallbacks by @tavren27 in #51836
  • fix(scim): answer every /scim/v2 error with the RFC 7644 SCIM Error schema by @tavren27 in #51322
  • test(spend): compare FOCUS RequesterDetails alias exactly after dropped-connection recovery by @devin-ai-integration[bot] in #52198
  • fix(logging): reuse the cached GenericAPILogger so config reloads stop leaking flush tasks by @Jeongho0805 in #50879
  • fix(cost): price forwarded-credential clones by their source deployment by @kerry-berri in #52313
  • fix(proxy): leave cache-read tokens out of the priority limiter's token count by @kenneth-berri in #52209
  • fix(openai): bill the audio output tokens OpenAI reports for gpt-4o-mini-tts speech by @kerry-berri in #52317
  • chore(cost-map): rolling registry update, fal_ai kling v4, azure gpt-6-luna flex, snowflake, bedrock/mantle claude 5.5 by @devin-ai-integration[bot] in #52299
  • test(integration): stop requiring safeguards calls to wait for the anthropic beta map by @yuneng-berri in #52329
  • fix(cost-map): bill gpt-4o-transcribe audio input at the rate OpenAI charges by @kerry-berri in #52303
  • feat(shadow-eval): add optional rubric grades by @joshua-berri in #52321
  • fix(gateway): accept x-api-key on inference routes for Anthropic SDK clients by @zoveki in #49410
  • fix(dashscope): bill 1h cache writes at the 1h rate by @kenneth-berri in #52331
  • fix(dashscope): add cache read and write prices for qwen4-max by @kenneth-berri in #52334
  • feat(cost-map): add bedrock bare, global and us rows for openai gpt-6.2-astra and gpt-6.2-sol by @devin-ai-integration[bot] in #52332
  • fix(proxy): make credential reload sync linear in the number of credentials by @joshua-berri in #52305
  • chore(cost-map): add fal_ai HeyGen Avatar 6 and Microsoft MAI-Image-2.7 rows by @devin-ai-integration[bot] in #52340
  • feat(proxy)!: require a LITELLM_SALT_KEY of at least 112 bits to start in fips_mode by @yucheng-berri in #52099
  • ci(osv): ignore picomatch GHSA-fc5q-39xc-vjpf for 7 days by @yuneng-berri in #52337
  • fix(redis): drop ssl_ kwargs from the async pool when ssl is off by @kerry-berri in #52344
  • chore(cost-map): add github_copilot mai-code-2 and mai-code-2-flash by @devin-ai-integration[bot] in #52338
  • fix(anthropic): accept temperature=1 on Claude models that removed sampling params by @kerry-berri in #52341
  • test(cost): add azure gpt flex service tier cost tracking cases by @devin-ai-integration[bot] in #50634
  • fix(proxy): reject comma-separated models with stream=true as 400 by @ryan-crabbe-berri in #52264
  • fix(mcp): use the saved description and server name in registry.json by @jesus-berri in #52319
  • test(database): give the autorouter baseline Prisma client the rig's startup budget by @devin-ai-integration[bot] in #52364
  • fix(responses): replay phase on the azure route and tag streamed text deltas with phase by @zoveki in #51059
  • feat(helm): allow labels and annotations on the gateway ConfigMap by @kirosvi in #44061
  • fix(anthropic): close the Responses upstream when a bridged /v1/messages stream is closed by @zoveki in #51274
  • fix(proxy/auth): return 401 for a non-master key when no database is configured by @kenneth-berri in #52349
  • feat(model_capacity): replace the python dynamic rate limiter v3 with a rust callback by @narmstrong-berri in #51540
  • chore(cost-map): add bedrock_mantle GPT-6.2 Sol and Astra rows by @DucNgn in #52359
  • refactor(rust): remove unused MCP, UI and management rust routes by @yujonglee-berri in #52221
  • fix(otel): emit a span for every failed retry and fallback attempt by @shivamrawat1 in #50310
  • fix(proxy): apply a JWT key mapping without jwt_issuer right away when issuers are configured by @kerry-berri in #52350
  • fix(sdk): keep proxy-only imports off Messages streaming and cache-hit paths by @kerry-berri in #52339
  • test(cost): mark anthropic cases for the cost tracking reconciliation automation by @kerry-berri in #52358
  • fix(grayswan): honor failure policy for invalid monitor decisions by @yucheng-berri in #52191
  • fix(auth): index restricted end-user registry refreshes by @ryan-crabbe-berri in #51772
  • refactor(ui): polish customer management portal by @moe-berri in #52381
  • test(management): isolate global policy tests from the shared peer proxy by @devin-ai-integration[bot] in #52318
  • feat(sarvam): send a litellm User-Agent on chat requests by @vikashmaddi-sarvam in #52012
  • fix(router): free streamed Responses API requests without waiting for a full gc by @ryan-crabbe-berri in #52215
  • perf(caching): hash response cache keys piece by piece and store under the looked-up key by @ryan-crabbe-berri in #52267
  • fix(gemini): keep Google's reason on streamed chunks and the assembled response by @ryan-crabbe-berri in #52242
  • feat(gateway): resolve Redis and the database in their own crates and hand them to routes as axum state by @yujonglee-berri in #52222
  • test(authorization): expect the sse stream_format the proxy now sends for speech billing by @devin-ai-integration[bot] in #52392
  • feat(proxy): page team members and stop loading full rosters for /team/info callers by @jesus-berri in #51885
  • fix(proxy): model_max_budget entry on a model group covers its global aliases by @kenneth-berri in #52385
  • fix(proxy): never wait on a tokenizer download when estimating a failed request's input tokens by @devin-ai-integration[bot] in #52232
  • chore(cost-map): add github_copilot gpt-6.2-astra and gpt-6.2-sol by @devin-ai-integration[bot] in #52395
  • feat(router): add congestion-triggered priority queue controller by @joshua-berri in #51883
  • fix(responses): prevent duplicate authorization headers by @AaronHowell in #52259
  • fix(aiohttp_transport): close sessions of evicted Azure and base64 embedding clients by @ryan-crabbe-berri in #52266
  • fix(github_copilot): send the Copilot CLI integration id for a custom GitHub App by @mateo-berri in #52386
  • fix(tokenizer): match tiktoken on lone surrogates and out-of-u32 tokens by @xinglan163 in #51720
  • feat(ui): take projects out of beta in the nav by @tin-berri in #52389
  • fix(guardrails): stop the alice guardrail sending each prompt four times by @seanyasno-af in #51392
  • fix(cli): carry configured service tiers and display names into the lite codex catalog by @zoveki in #49422
  • fix(mcp): decide connected from the provider's scope refusal, not from scope names by @shrey-berri in #51788
  • fix(github_copilot): explain why a Copilot call failed and how to fix it by @mateo-berri in #52352
  • test(providers): cover the prompt-management marker and vispark in the internal params wire walk by @devin-ai-integration[bot] in #52363
  • test(mcp): give the startup-failure child process the rig's readiness budget by @devin-ai-integration[bot] in #52365
  • feat(nvidia_nim): refresh the catalog and discover upstream models by @mateo-berri in #50439
  • test(spend): snapshot logs after the denied request is recorded by @devin-ai-integration[bot] in #52422
  • test(database): expect the restricted end-user index build by @devin-ai-integration[bot] in #52421
  • feat(fips): encrypt stored secrets with the FIPS provider's AES-256-GCM in fips_mode by @yucheng-berri in #51429
  • fix(decisions): forward OpenAI prompt cache fields by @nicolasmota in #51457
  • feat(router): coordinate congestion queues through Redis by @joshua-berri in #52367
  • fix(auth): check a mapped JWT's end-user claim and keep the error classes of refused keys and JWTs by @yujonglee-berri in #52397
  • fix(bedrock): stop the prefill 400 when the last user message is empty by @zoveki in #50701
  • fix(google_genai): honor snake_case generation_config on generateContent by @mika-lt in #51139
  • feat(auth): read virtual keys, jwt key mappings and model access groups through rust repositories by @narmstrong-berri in #52173
  • feat(jwt): map customer org, customer member and customer team from JWT claims by @ishaan-j-berri in #52399
  • feat(ui)!: remove Projects toggle and align Enterprise access by @ryan-crabbe-berri in #49391
  • test(auth): back mapped jwt deployment profile cases with rust repository fakes by @narmstrong-berri in #52442
  • fix(proxy): run the empty models boot counts one at a time so workers don't open six Postgres connections by @kerry-berri in #52378
  • test(perf): large-body 400-user case and mixed-path memory soak for escaped regressions by @kerry-berri in #52404
  • test(perf): recalibrate the overhead p50 ceilings after the #47790 header fix by @kerry-berri in #52410
  • fix(e2e): verify Projects license notice and page content by @tin-berri in #52454
  • fix(proxy): require proxy admin role on GET /customer/info by @kerry-berri in #52451
  • feat(router): queue congested provider requests by priority by @joshua-berri in #51892
  • fix(streaming): report upstream failures after client disconnect by @mateo-berri in #52450
  • test(perf): give responses_2rps its own overhead p50 ceiling by @kerry-berri in #52459
  • fix(azure_ai)!: send chat on Foundry project and /openai/v1 bases to the v1 chat route by @mika-lt in #51144
  • fix(cost-map): bedrock gpt-6.2 astra mantle ultrafast prices and model card sources for gpt-6.2 astra and sol by @devin-ai-integration[bot] in #52464
  • fix(streaming): preserve failure context without retaining requests by @mateo-berri in #52408
  • test(management): size team delete lock probes for the Prisma pool split by @devin-ai-integration[bot] in #52481
  • test(cost): add gemini-3.1-pro-preview flex tier cost tracking case by @devin-ai-integration[bot] in #47845
  • fix(responses): preserve single-object input through the completion bridge by @devin-ai-integration[bot] in #52483
  • feat(customer-portal): reuse budget groups for organizations by @tin-berri in #52411
  • test(auth): size the concurrent first JWT call proxy for the Prisma pool split by @devin-ai-integration[bot] in #52486
  • docs(review): require the breaking-change marker only for evidenced breaks and live proof for LLM translation PRs by @mateo-berri in #52465
  • chore(cost-map): add together_ai/Qwen/Qwen4-Max from the Together models API by @devin-ai-integration[bot] in #52487
  • refactor: clean up fresh tech debt from 2026-11-18 by @devin-ai-integration[bot] in #52491
  • fix(proxy): resolve a team wildcard deployment for a wildcard key on files, batches and RAG routes by @tavren27 in #49744
  • feat(providers): add Sarvam Token Factory v2 models by @vikashmaddi-sarvam in #49792
  • fix(responses): stop the background stream when a polling response is cancelled by @tavren27 in #49384
  • chore(cost-map): fill Bedrock regional and Azure Government pricing gaps, rolling registry audit 2026-11-18 by @devin-ai-integration[bot] in #52458
  • chore(cost-map): add AWS Bedrock Mantle GovCloud Nemotron Nano deprecation dates by @devin-ai-integration[bot] in #52552
  • fix(cost): bill OpenRouter and Perplexity v1.1 decisions models by @devin-ai-integration[bot] in #45476
  • test(cost): add tier 1 cost tracking cases for gpt-6.1-terra and sync suite price fields by @devin-ai-integration[bot] in #49587
  • fix(cost): bill web and file search together by @kerry-berri in #52555
  • chore(cost-map): drop claude-mythos-preview deprecation_date and add azure/us-gov/gpt-4o-mini by @devin-ai-integration[bot] in #52557
  • feat(moonshot): add Kimi K3 and Together AI Inkling model support by @devin-ai-integration[bot] in #33922
  • fix(azure_ai)!: call Responses natively for a model router on a Foundry project or resource base by @zoveki in #52474
  • feat(proxy): configure customer identity headers by @ishaan-j-berri in #52467
  • feat(vector_stores): add YAML-configured unified search fallbacks by @mrinal-berri in #52093
  • feat(proxy): add a surcharge on customer requests by @ishaan-j-berri in #52406
  • fix(ui): recognize Bedrock ca. and in. inference profiles in auto router setup by @kerry-berri in #52574
  • feat(ui): offer Lens in the workspace switcher in place of Moyai by @kerry-berri in #52563
  • feat(customer portal): set budgets per customer and gateway app by @ishaan-j-berri in #52477
  • feat(ui): expose customer and app TPM/RPM budget limits by @tin-berri in #52493
  • ci: run a core set of unit, lint and smoke jobs on pull requests and the rest on main by @kerry-berri in #52582
  • fix(integration): size rig pools for the rust connection share and stop the router importing the proxy by @narmstrong-berri in #52532
  • fix(ui): mark logging callbacks set in config.yaml and disable their delete by @yucheng-berri in #52507
  • ci: split the core router unit job into three by @kerry-berri in #52594
  • test(scim): give the group delete lock test two Prisma connections by @devin-ai-integration[bot] in #52607
  • fix(groq): restore Whisper transcription pricing by @kerry-berri in #52570
  • fix(router): do not retry a Responses stream after an output item reached the client by @kerry-berri in #52461
  • fix(cost-map): set chatgpt/gpt-5.6 max_input_tokens to 922000 to match the openai rows by @devin-ai-integration[bot] in #52614
  • feat(proxy): let each app decide whether it may create unknown customers by @moe-berri in #52485
  • fix(proxy): apply a view-only Default Role to users with no stored role by @Ayushshuklai in #52295
  • fix(logging): stop keeping a full response copy nobody reads by @ryan-crabbe-berri in #52524
  • feat(proxy): let an app set its own surcharge on customer requests by @moe-berri in #52492
  • fix(logging): redact generated images when message logging is turned off by @kenneth-berri in #52605
  • fix(proxy): recover from a read replica reconnect that times out by @kenneth-berri in #52592
  • feat(ui): list every application on the Applications sidebar page by @moe-berri in #52496
  • feat(ui): improve customer budget editing and spend tables by @tin-berri in #52608
  • feat(model_armor): scan MCP tool results on post_mcp_call by @jesus-berri in #51887
  • fix(datadog): send DD_TAGS on Datadog logs and LLM Observability spans by @yucheng-berri in #52415
  • feat(cost-map): add gemini/gemini-4-neon by @devin-ai-integration[bot] in #52640
  • fix(anthropic, bedrock): keep adaptive thinking when tool-call history has no thinking blocks by @kenneth-berri in #52615
  • fix(decisions): match OpenAI format answers to questions by name by @kenneth-berri in #52597
  • fix(tavily): price the search depth chosen automatically by @kerry-berri in #52564
  • fix(bedrock): bill ConverseStream at the served tier by @kerry-berri in #52565
  • fix(gemini): bill cached input tokens on image generation by @kerry-berri in #52572
  • fix(cost): bill SVG authoring usage on OpenAI image responses by @kerry-berri in #52571
  • fix(gateway-forward): keep x-litellm-api-key when the client names it in connection by @narmstrong-berri in #52427
  • refactor(router): delete router code nothing calls by @yujonglee-berri in #52516
  • fix(responses): send bridged system prompts as input messages so GPT-6.2 caches them by @mateo-berri in #52599
  • fix(groq): default service_tier when the reply omits it by @narmstrong-berri in #52431
  • docs(review): require live QA across shared translation routes by @mateo-berri in #52631
  • feat(cost-map): add vertex_ai/gemini-4-neon by @devin-ai-integration[bot] in #52648
  • test(configuration): initialize JSON startup databases before capture by @devin-ai-integration[bot] in #52650
  • fix(responses): prevent stream restarts after non-text output by @kerry-berri in #52616
  • test(pricing): derive transcription charges from the cost map by @devin-ai-integration[bot] in #52653
  • test(management): isolate the seeded roster per test by @devin-ai-integration[bot] in #52652
  • fix(harness): surface Pi context compaction events by @moe-berri in #52655
  • fix(db): fail rust pool acquires at once on a refusing postgres, with #52532 integration follow-ups by @narmstrong-berri in #52617
  • test(anthropic): cover adaptive thinking on tool turns under modify_params by @ryan-crabbe-berri in #52425
  • feat(db): rust repositories for user, end user and budget auth objects by @narmstrong-berri in #52170
  • feat(auth): read teams, team memberships and organizations through rust repositories by @narmstrong-berri in #52176
  • feat(auth): read projects, tags, access groups and object permissions through rust repositories by @narmstrong-berri in #52178
  • fix(gemini): send minimal thinking to Nano and drop settings image models refuse by @ryan-crabbe-berri in #52514
  • fix(responses): log the resolved provider on Responses API calls by @ryan-crabbe-berri in #52622
  • fix(docker): disable Prisma's update check so it stops leaving zombie processes by @kenneth-berri in #52642
  • fix(proxy): enforce a key's max_budget on custom auth requests by @kenneth-berri in #52654
  • fix(voyage): honor VOYAGE_API_BASE for embeddings by @kenneth-berri in #52645
  • ci: fit the pull request gate into five minutes by @yuneng-berri in #52660
  • test: stabilize the flaky prompt cache detector unit tests by @yuneng-berri in #52671
  • fix(guardrails): write rewritten tool_calls from the Generic Guardrail API back to the client by @jesus-berri in #50413
  • feat(proxy): report current budget-window spend on /user/info, /v2/user/info and /team/info by @mateo-berri in #52673
  • fix(cost-map): price fireworks_ai kimi-latest as Kimi K3.1 Pro by @devin-ai-integration[bot] in #52634
  • test(auth): seed fallback budgets in the Rust repository fake by @kerry-berri in #52685
  • perf(proxy): coalesce empty auto-router baseline recovery polls by @kerry-berri in #52523
  • feat(cost-map): add github_copilot/gemini-4-neon by @devin-ai-integration[bot] in #52704
  • chore(cost-map): keep GovCloud Mantle Claude token counting in us-gov-west-1 by @devin-ai-integration[bot] in #52708
  • test(prompt-templates): avoid exhausting recursion in regex coverage by @tin-berri in #52706
  • fix(azure-ai): send max_completion_tokens for GPT-5 Chat on Foundry by @moe-berri in #52680
  • test(responses): expect bridged system prompts as input messages in tests #52599 left stale by @mateo-berri in #52693
  • feat(cost-map): add Databricks GPT-6.2 Astra and Sol Unity Gateway rows by @devin-ai-integration[bot] in #52635
  • fix(ollama): log missing prompt_eval_count through verbose_logger by @narmstrong-berri in #52432
  • fix(bedrock): send text.format.strict for GPT native structured output on Converse by @ryan-crabbe-berri in #52419
  • test(integration): expect the system prompt as the first Responses input message by @devin-ai-integration[bot] in #52697
  • fix(proxy): cache customer app definition lookups by @narmstrong-berri in #52688
  • test(perf): recalibrate the small-body Postgres statement ceiling by @kerry-berri in #52686
  • fix(proxy): release request bodies of failed calls without waiting for a full GC by @kerry-berri in #52600
  • fix(proxy): count generateContent spend under its own endpoint by @ryan-crabbe-berri in #52701
  • fix(gemini): drop logprobs on every model Google refuses them for by @ryan-crabbe-berri in #52700
  • test(cost): add tier 1 cost tracking cases for gpt-6.2-sol, gpt-6.2-astra and azure_ai claude-opus-5-5 by @kerry-berri in #52587
  • fix(proxy): skip prisma generate in the migration job when the prisma package is not writable by @kerry-berri in #52460
  • feat(gateway): verify the LiteLLM license once at boot by @yujonglee-berri in #52227
  • fix(responses): scope cacheable system input to stateless requests by @mateo-berri in #52694
  • fix(ui): match Model Hub to the model catalog layout by @tin-berri in #52672
  • feat(messages): serve bedrock mantle messages on rust only by @maciej-berri in #52328
  • chore(cost-map): price Cohere free-tier Command A Reasoning, Translate and Vision at zero by @kerry-berri in #52687
  • chore(cost-map): rolling registry audit 2026-11-18, Together/xAI/DeepSeek flags, Cohere limits, OpenRouter, Jina and Perplexity rows by @devin-ai-integration[bot] in #52619
  • chore(cost-map): add fal_ai speridlabs iris-z depth estimation rows by @devin-ai-integration[bot] in #52715
  • ci(unit): split core-router and core-auth-spend shards by @kerry-berri in #52719
  • ci: stop running root entry limit on every PR by @kerry-berri in #52579
  • ci: reuse Rust bridge builds across PR pushes and refresh the Cargo cache by @kerry-berri in #52718
  • test(guardrails): give each Lakera outage burst call its own prompt by @yucheng-berri in #52730
  • ci(rust): run LiteLLM Rust DB on pushes to main only and cover release-line PRs by @kerry-berri in #52609
  • fix(cost-map): move azure/gpt-image-1.5 retirement to 2027-06-16 and add Codestral-2501 successor by @devin-ai-integration[bot] in #52699
  • chore(model_prices): fill successor_model and deprecation_source for dated models by @kerry-berri in #52710
  • fix(chatgpt): preserve parallel_tool_calls and enforce Codex Responses-Lite contract by @kenneth-berri in #52731
  • ci: skip unit-core shards on test-only PRs by @kerry-berri in #52720
  • fix(logging): keep the resolved provider in both logged fields by @ryan-crabbe-berri in #52709
  • fix(sap): accept list-shaped reasoning_content from Gemini 3.x by @kenneth-berri in #52741
  • test(bedrock): expect the fixed-sampling temperature message from #52341 in the Mantle wire test by @kerry-berri in #52742
  • fix(proxy): keep stored values under LITELLM_SALT_KEY when rotating the master key by @yucheng-berri in #52413
  • chore: bump litellm-enterprise 0.1.86 -> 0.1.87, litellm-proxy-extras 0.4.117 -> 0.4.119, litellm 1.111.0 -> 1.112.0 by @yuneng-berri in #52748
  • feat(customer-portal): show token usage and cache costs by @ishaan-j-berri in #52601
  • test(proxy): drain queued router accounting before comparing TPM sequences by @kerry-berri in #52750
  • test(router): pin the clock in the usage-based-v2 RPM pre-call test by @kerry-berri in #52746
  • feat(gateway): serve one native app in both modes, answering a master key caller's /v1/messages with it in front of Python by @yujonglee-berri in #52229
  • feat(proxy)!: let every key read its own spend logs and return 403 for filters outside the caller's scope by @kerry-berri in #52449
  • test(auth): use the production cache in auth fixtures by @devin-ai-integration[bot] in #52754
  • feat(gateway): port the proxy's middleware onto the native app by @yujonglee-berri in #52223
  • refactor(router-core): hand the host to each routing call instead of keeping it on the router by @yujonglee-berri in #52509
  • fix(router): fall back by the group a pick was routed to, whatever fails after the reroute by @yujonglee-berri in #52510
  • fix(responses): ignore dropped Messages placement parameters by @mateo-berri in #52743
  • fix(gateway): honor the configured Postgres connection cap by @kerry-berri in #52729
  • fix(cost-map): price databricks system.ai.gpt-6-2-astra and gpt-6-2-sol from the Databricks DBU rates by @devin-ai-integration[bot] in #52762
  • fix(proxy): stop passing cache_params.default_redis_ttl to the redis client by @narmstrong-berri in #52440
  • test(responses): match stored requests in the shared wire fixture by @devin-ai-integration[bot] in #52765
  • fix(vertex_ai): stop a json_schema row failing the plain rows in its batch file by @zoveki in #51790
  • test(images): expect cache fields in image usage by @devin-ai-integration[bot] in #52777
  • feat(logging): add the StandardLoggingPayload types in rust by @yujonglee-berri in #52783
  • fix(ui): link env-credential login warning to the docs step that disables it by @mateo-berri in #52788
  • fix(ui): hide gateway sidebar pages on the Lens workspace by @kerry-berri in #52771
  • fix(tests): repair the unit tests failing on main's full run by @mateo-berri in #52769
  • fix(ui): add spacing below the Add Fallbacks button by @mateo-berri in #52793
  • feat(ui): follow the system color scheme when no theme is saved by @mateo-berri in #52792
  • feat(ui): show the new LiteLLM logo on the login page by @mateo-berri in #52789
  • revert: reuse Rust bridge builds across PR pushes and refresh the Cargo cache (#52718) by @kerry-berri in #52770
  • test(images): expect cache fields in gemini image edit usage by @kerry-berri in #52798
  • fix(cost-map): price together_ai/Qwen/Qwen4-Max at the promotional rate Together bills until 2026-11-23 by @devin-ai-integration[bot] in #52814
  • test(azure): restore client session after extra headers test by @devin-ai-integration[bot] in #52815
  • refactor: clean up fresh tech debt in the provider config loader and prompt helpers by @devin-ai-integration[bot] in #52817
  • fix(guardrails): mask reasoning text on /v1/responses output by @quietjules in #52504
  • fix(proxy): send no body upstream on a pass-through GET or DELETE sent without one by @quietjules in #50833
  • refactor: clean up fresh tech debt in routing, riva, least-busy, proxy cache and prompt template helpers by @mateo-berri in #50192
  • feat(config): read env-backed litellm_settings keys from LITELLM_ under config.yaml by @yujonglee-berri in #52822
  • refactor(messages): name the Messages body fields once, on the request type by @yujonglee-berri in #52836
  • refactor(framer): relay SSE blocks without dropping a byte by @yujonglee-berri in #52544
  • refactor(router-types): move RoutingStrategy and RoutingGroupSpec out of router-core by @yujonglee-berri in #52847
  • fix(cost-map): refresh Azure and Together retirement dates and add GMI Gemini 4 Hydrogen by @devin-ai-integration[bot] in #52763
  • fix(mcp): return an upstream -32042 URL elicitation error from /mcp as a JSON-RPC error by @jesus-berri in #52384
  • fix(key-management): stop ownerless keys from reading each other's key info by @ojensen-berri in #52838
  • feat(cost-map): add azure_ai/kimi-k3.1-pro at the Azure Global Standard rate by @kerry-berri in #52873
  • fix(proxy): prune expired LiteLLM_SpendLogGuardrailIndex rows with spend-log retention by @jesus-berri in #51518
  • fix(proxy-extras): keep the host query param in the v2 migration lock URL by @yucheng-berri in #52797
  • feat(cost-map): add gemini/genie-4 at the prices Google published on 2026-11-19 by @devin-ai-integration[bot] in #52872
  • fix(router): forward Responses lifecycle events after output instead of dropping them on stream failure by @kerry-berri in #52858
  • fix(images): read null cached token counts as 0 in image usage by @kerry-berri in #52857
  • fix(bedrock): return a response for empty Invoke streams with complete_response by @kerry-berri in #52862
  • fix(logging): restore exception in model group rate limit callback kwargs by @kerry-berri in #52861
  • fix(bedrock): bill legacy bedrock/mantle GovCloud calls at the region's rates by @kerry-berri in #52864
  • feat(vector-stores): fail over native search passthrough to configured fallback stores by @mrinal-berri in #52856
  • fix(images): honor additional_drop_params given as a tuple or holding non-string entries by @kerry-berri in #52390
  • fix(perf): integrate pod CPU windows so cpu_seconds_per_request stops under-reading by @kerry-berri in #52735
  • fix(auth): keep refusing other exhausted limits before the body when the team is not cached by @kerry-berri in #52388
  • test(images): include cache and SVG fields in image routing fixture by @devin-ai-integration[bot] in #52898
  • feat(proxy): app-scoped read access to Customer Management through object permissions by @moe-berri in #52781
  • fix(customer-management)!: preserve permanent budgets across temporary increases by @tin-berri in #52787
  • test(utils): skip gemini/genie-4 in the gemini tpm/rpm check by @shrey-berri in #52900
  • chore(model_prices): fill remaining successor_model for dated models by @devin-ai-integration[bot] in #52893
  • test(translation): align stored Responses bridge fixtures by @devin-ai-integration[bot] in #52894
  • fix(decisions): support the litellm_proxy provider by @bofenghuang in #51611
  • fix(httpx): repair surrogate code points in raw text request bodies by @anencore94 in #49599
  • chore(docs): regenerate provider endpoints support by @yuneng-berri in #52918
  • fix(caching): allow excluding embeddings from response caching by @joshua-berri in #52902
  • feat(customer-portal): unify budget editors and usage tables by @tin-berri in #52682
  • fix(embeddings): preserve custom HTTP client error statuses by @joshua-berri in #52904
  • ci(unit): run the proxy-extras tests when a pull request touches proxy-extras by @yuneng-berri in #52917
  • docs(agents): prove translation and cost fixes on every request shape by @mateo-berri in #52911
  • ci(lint): clear the two code-quality failures behind the stale provider file by @yuneng-berri in #52921
  • fix(responses): keep Claude Code billing metadata out of prompts by @mateo-berri in #52732
  • feat(proxy): run callback setup and cleanup hooks during the proxy lifespan by @kenneth-berri in #52908
  • feat(cost-map): add openai gpt-live-1.1 at the per-minute rate OpenAI published on 2026-11-19 by @devin-ai-integration[bot] in #52926
  • fix(cost-map): lower four fireworks_ai spot router rows to the prices the serverless models API now lists by @devin-ai-integration[bot] in #52928
  • fix(router): scope cooldown reads to request candidates by @joshua-berri in #52913
  • chore(cost-map): add fal_ai Ideogram 4.6 text to image and edit rows by @devin-ai-integration[bot] in #52927
  • fix(auth): bound model access group registry scans by @joshua-berri in #52899
  • fix(proxy-extras): make the app association budget migration re-runnable by @yuneng-berri in #52909
  • fix(key-management): let a key read its own info by its plaintext key by @yuneng-berri in #52925
  • test(s3-log): wait for the async upload failure report before checking it by @narmstrong-berri in #52737
  • fix(guardrails): read per-request extra_body from the proxy request in the generic guardrail api by @narmstrong-berri in #52438
  • feat(messages): serve vertex ai claude messages on rust only by @maciej-berri in #52314
  • feat(db): rust repository for the spend logs window spend sum by @narmstrong-berri in #52661
  • test(sdk): drain streaming logs before the next fallback test by @devin-ai-integration[bot] in #52945
  • test(spend): include cache and margin fields in the daily activity repository goldens by @devin-ai-integration[bot] in #52944
  • fix(vertex_ai): count thinking tokens on non-streamed content filter replies by @kenneth-berri in #52930
  • fix(router): optionally reject raw model collisions across groups by @joshua-berri in #52895
  • fix(proxy): keep pass-through handler usage when finalizing cost before dispatch by @chuckiihub in #52527
  • fix(responses): send an HTTP status on managed websocket error frames by @ryan-crabbe-berri in #52874
  • fix(responses): cache bridged system prompts on stored GPT-6.2 requests by @mateo-berri in #52906
  • feat(guardrails): opt-in pre_call scan of tool descriptions as text for PANW and Presidio by @mrinal-berri in #50301
  • refactor(ui): remove unused agentSnippets from model hub by @yuneng-berri in #52931
  • refactor(ui): remove unused getCategories from compliance prompts by @yuneng-berri in #52932
  • refactor(ui): remove unused EndpointField from model hub parts by @yuneng-berri in #52933
  • fix(proxy): org admin spend log scope and gemini file id parsing by @kerry-berri in #52943
  • fix(auth-gcp): build the key file version on Windows by @yuneng-berri in #52951
  • fix(cost-map): halve the fireworks_ai deepseek-v4p1-flash-1017 ultrafast price and add its flex router id by @kerry-berri in #52947
  • fix(responses): send an HTTP status on websocket guardrail block frames by @ryan-crabbe-berri in #52942
  • feat(messages): serve bedrock claude_platform messages on rust only by @maciej-berri in #52681
  • fix(spend): record the bound on the org admin spend log filter by @yuneng-berri in #52961
  • test(secret-detection): keep the linear-scan test stable when the re cache is evicted by @yucheng-berri in #52940
  • refactor(auth-types): move the Claude Platform params next to AwsParams and VertexParams by @maciej-berri in #52966
  • feat(anthropic): price Claude subscription OAuth allowance usage at $0, overage at API rates by @kerry-berri in #52901
  • fix(packaging): ship the license public key in the litellm-proxy sdist by @yuneng-berri in #52970

New Contributors

Full Changelog: v1.111.0-dev.4...v1.112.0-dev.1