BerriAI/litellm v1.91.0-dev.2 on GitHub

Verify Docker Image Signature

All LiteLLM Docker images are signed with cosign. Every release is signed with the same key introduced in commit 0112e53.

Verify using the pinned commit hash (recommended):

A commit hash is cryptographically immutable, so this is the strongest way to ensure you are using the original signing key:

cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/0112e53046018d726492c814b3644b7d376029d0/cosign.pub \
  ghcr.io/berriai/litellm:v1.91.0-dev.2

Verify using the release tag (convenience):

Tags are protected in this repository and resolve to the same key. This option is easier to read but relies on tag protection rules:

cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/v1.91.0-dev.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.91.0-dev.2

Expected output:

The following checks were performed on each of these signatures:
  - The cosign claims were validated
  - The signatures were verified against the specified public key

What's Changed

fix(bedrock-mantle): honor api_base for VPC endpoint routing on bedro… by @shivamrawat1 in #31141
fix(docker): bump wolfi-base digest to patch openssl CVE-2026-34182 by @yucheng-berri in #31133
feat: add minimal rust router + axum ai-gateway calling router.realtime (2/2) by @ishaan-berri in #31135
build: add Dockerfile + render blueprint for rust ai-gateway by @ishaan-berri in #31154
fix(mcp): resolve config-defined servers in per-user credential and env-var endpoints by @tin-berri in #31171
perf: pre-warm upstream realtime connection pool to cut session-establishment latency by @ishaan-berri in #31163
chore: clarify rule about trailing periods by @mateo-berri in #31175
fix(anthropic): drop unsupported speed param with drop_params by @krrish-berri-2 in #31152
fix(proxy): expand all-proxy-models sentinel in direct access lookup by @mubashir1osmani in #31153
fix(anthropic): sanitize tool_use ids on native /v1/messages path by @Sameerlite in #31094
fix(ui): persist budget window deletion on virtual keys by @ryan-crabbe-berri in #31107
feat(ui): track frontend lint counts in a committed snapshot by @ryan-crabbe-berri in #31157
refactor(litellm-rust): dissolve providers into core + ai-gateway (strict 3-crate layers) by @ishaan-berri in #31218
fix(vertex/files): stream OpenAI->Vertex batch JSONL uploads by @mubashir1osmani in #31036
fix(mcp): let proxy admins assign MCP servers to teamless keys by @tin-berri in #31126
feat(mcp): graft v2 resolver onto _create_mcp_client (none + api_key static family) by @tin-berri in #31058
test: add e2e tests for spend, budgets and llms by @mateo-berri in #30869
feat(proxy): add POST /v1/callbacks/logs to replay logging payloads through callbacks by @ishaan-berri in #31134
fix(ui): render logos under a custom server_root_path by @ryan-crabbe-berri in #31156
feat: make rust OCR async-first by @ishaan-berri in #31253
test(e2e): drop xfail markers for now-fixed team-budget-JSON and custom-pricing-leak bugs by @mubashir1osmani in #31249
fix(proxy): stop double-decrypting email/slack alerting env vars in get_config by @mubashir1osmani in #31117
fix(mcp): correct misleading no-trusted-proxy warning for XFF access control by @mateo-berri in #31264
fix(mcp): warn loudly when X-Forwarded-For is present but use_x_forwarded_for is off by @mateo-berri in #31266
fix(mcp): resolve toolset tools by the server's known prefix by @tin-berri in #31254
fix(mcp): challenge delegate-auth OAuth servers with upstream resource_metadata by @tin-berri in #31255
fix(ci): point OSS contributor workflows to litellm_oss_staging by @Sameerlite in #31270
fix(streaming): word-sliced cache replay for stream=true cache hits by @michelligabriele in #30216
feat(mcp): add mcp_xff_num_trusted_hops to harden X-Forwarded-For client IP resolution by @mateo-berri in #31257
chore: migrate Python formatter from black to ruff format by @mateo-berri in #31317
fix(otel): hashable scope for _emit_once when guardrail_mode is list by @yucheng-berri in #31262
feat: package Rust OCR bridge in LiteLLM wheel by @ishaan-berri in #31267
chore: gitignore rust bridge build artifacts by @mateo-berri in #31349
ci: harden cargo fetches during maturin builds by @ishaan-berri in #31348
chore(lint): widen ANN slack to 10% of baseline and drop PLR0913 from the strict gate by @mateo-berri in #31335
feat: add Rust OCR providers by @ishaan-berri in #31272
fix(cache): apply Redis namespace to all key operations by @yassin-berriai in #31288
feat(pricing): add gemini-3-pro-image and gemini-3.1-flash-image GA model pricing by @milan-berri in #30022
fix: clarify further that customer names shouldn't be made public by @mateo-berri in #31365
ci(image-scan): add Grype image scan for OS + library CVEs by @yucheng-berri in #31151
feat(aiml): add openai/gpt-image-2 image model by @mateo-berri in #31323
feat(mistral): support Mistral OCR 4 (mistral-ocr-4-0) by @mateo-berri in #31353
fix: inverted rule in CLAUDE.md by @mateo-berri in #31370
fix(proxy/client): redact api key from key/info client error messages by @yucheng-berri in #31342
feat(spend): store litellm_call_id on spend logs for DB-to-trace correlation by @yucheng-berri in #31344
chore(deps): bump deps by @yuneng-berri in #31377
fix(cost-map): retarget mistral-medium-latest to Medium 3.5 and add date-pinned aliases by @mateo-berri in #31373
feat(ocr): thin Rust OCR Python bridge by @ishaan-berri in #31368
feat(mcp): opt-in least-privilege default for team key MCP access by @ryan-crabbe-berri in #31380
chore(ci): main into internal_staging (reconcile OCR hotfix history; unblocks #31384) by @yuneng-berri in #31390
fix(vertex): preserve Gemini Embedding 2 usageMetadata for cost tracking by @mateo-berri in #31354
chore: remove CI section from PR template by @mateo-berri in #31376
test(logging): regression coverage for streaming /v1/messages OpenAI Responses spend logs by @yucheng-berri in #31388
chore: litellm oss staging 250626 by @Sameerlite in #31305
fix(vertex_ai): prevent stale Vertex bearer token causing /v1/messages 401 after token expiry by @Sameerlite in #31276
fix(proxy): skip model override when response has no model field by @Sameerlite in #31183
fix(vertex): stop O(n^2) re-parse of accumulated Gemini stream JSON by @yassin-berriai in #31297
fix(router): surface clean RateLimitError on mid-stream 429 with no fallbacks by @yassin-berriai in #31298
build(docker): build the Admin UI from source in a build-platform-pinned stage by @tin-berri in #31130
fix(bedrock_guardrails): select latest user message by original role in apply_guardrail by @michelligabriele in #30482
fix(proxy): restore wildcard expansion in /v1/model/info by @Sameerlite in #31444
fix(cli): mint per-session agent credential on lite login by @Sameerlite in #31072
chore: litellm oss staging by @Sameerlite in #31185
fix(cost): restore per-query Gemini 3.x web search billing by @mateo-berri in #31363
chore(ci): reconcile main into internal_staging to unblock promotion (#31384) by @yuneng-berri in #31392
test: add realtime proxy e2e suite across providers by @Sameerlite in #30960
perf(spend-logs): only strip NUL bytes in safe_dumps when present by @yassin-berriai in #31424
perf(cost-calc): precompute service-tier cost-key suffixes by @yassin-berriai in #31431
perf(caching): memoize _get_all_llm_api_params, rebuilt per request by @yassin-berriai in #31430
chore(ci): promote internal staging to main by @yuneng-berri in #31384
feat(mistral): add mistral/mistral-ocr-2512 (OCR 3) to cost map by @mateo-berri in #31463
feat(guardrails): make the Generic Guardrail resilient to built-in tools and errors (adopted from #31286) by @yucheng-berri in #31461
test(pass-through): fix langfuse auth=true test broken by allowed_passthrough_routes gate by @yuneng-berri in #31420
fix(team): persist budget_duration on /team/member_add member budgets by @yassin-berriai in #31443
fix(spend): fold logs-tab total into the page query to avoid a separate COUNT(*) by @yassin-berriai in #31423
fix(bedrock): surface web identity token aud/iss on InvalidIdentityToken by @yassin-berriai in #31412
fix(guardrails): instrument during-call and post-call guardrail latency by @yassin-berriai in #31414
fix(auth): cache auth-path team object under canonical team_id key by @yassin-berriai in #31418
fix(build): restore pure-Python uv_build backend to unblock PyPI publish by @yuneng-berri in #31470
chore(ci): promote internal staging to main by @yuneng-berri in #31477

Full Changelog: v1.91.0-dev.1...v1.91.0-dev.2