BerriAI/litellm v1.106.0-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:v1.106.0-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/v1.106.0-dev.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.106.0-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

refactor: expose public hidden_params accessors by @devin-ai-integration[bot] in #44668
fix(azure): date claude-opus-5-5 and sonnet-5-5 and fill azure/eu/gpt-6-astra limits by @berriai-litellm-provider-info-sync[bot] in #45018
refactor: remove fresh tech debt from the 2026-10-06 window by @devin-ai-integration[bot] in #45017
ci: give the database-backed CircleCI jobs their own Postgres by @devin-ai-integration[bot] in #45007
test(proxy): make the stagger-offset and linear-dedup guards independent of runner identity and load by @devin-ai-integration[bot] in #45031
refactor(types): replace Any with proven types in 16 files by @devin-ai-integration[bot] in #45029
fix(auth): clear the recent-miss user memo when /user/new creates the user by @devin-ai-integration[bot] in #45020
fix(health): attribute background health check results to their own deployment by @devin-ai-integration[bot] in #44982
fix(ollama): turn streamed prompt-based JSON tool calls into real tool calls by @devin-ai-integration[bot] in #45053
fix(bedrock): update GovCloud OpenAI prices, add GPT-6 Astra ultrafast tier and Titan Image v2 EOL by @berriai-litellm-provider-info-sync[bot] in #45077
fix(cost-map): add us data residency multiplier to anthropic claude-opus-5-5 by @berriai-litellm-provider-info-sync[bot] in #45078
refactor(llms): expose public names for private provider helpers by @devin-ai-integration[bot] in #45037
fix(bedrock): add 2027-03-30 deprecation date to DeepSeek R1 and Qwen3 Coder rows by @berriai-litellm-provider-info-sync[bot] in #45089
fix(proxy): opt-in Redis hash-tag grouping for v3 rate limiter by @devin-ai-integration[bot] in #45085
fix(bedrock): take context and output limits from the Bedrock model cards by @berriai-litellm-provider-info-sync[bot] in #45091
fix(tests): match the lowercased bind error in the owned-proxy port-race retry by @devin-ai-integration[bot] in #45097
test(e2e): tag llm_translation tests with Subject metadata and record harness steps by @ryan-crabbe-berri in #44950
test(e2e): tag claude_code cells with Subject metadata and record CLI driver steps by @ryan-crabbe-berri in #44961
test(e2e): tag management tests with Subject metadata and record management client steps by @ryan-crabbe-berri in #44962
test(e2e): tag guardrails and logging tests with Subject metadata and record client steps by @ryan-crabbe-berri in #44963
fix(caching): skip the cache past max_messages and keep tool_result text in semantic prompts by @devin-ai-integration[bot] in #43878
test(e2e): tag router, batches and mcp tests with Subject metadata and record client steps by @ryan-crabbe-berri in #44964
test(e2e): tag the remaining quota_management tests and record budget and spend client steps by @ryan-crabbe-berri in #44966
fix(router): preserve native baseline identity and accounting by @tin-berri in #44960
test(integration): scope the team-scoped models upstream check to its own model by @devin-ai-integration[bot] in #45106
fix(prometheus): add model_group label to end-to-end latency metrics by @ahamedshaik16 in #44860
test(mcp): fix stale bridge-hook, applied-guardrails and pagination-revoke integration tests by @devin-ai-integration[bot] in #45008
test(observability): wait for warm-up spend rows in capped and count only the post-wipe half in X4 by @devin-ai-integration[bot] in #45006
fix(cost-map): lower anthropic claude-sonnet-5-5 cache read price by @berriai-litellm-provider-info-sync[bot] in #45113
fix(prompt-caching): default injected /v1/messages breakpoints to implicit lookup by @ryan-crabbe-berri in #45121
test(integration): turn the model-info refresh off in every integration proxy config by @devin-ai-integration[bot] in #45104
test(e2e): let the realtime send step accept input audio buffer frames by @ryan-crabbe-berri in #45127
feat(model_prices): add claude-haiku-5-5 model pricing by @krrish-berri-2 in #45108
fix(realtime): run transcript guardrails on raw-path transcription sessions with a transcription-safe block by @devin-ai-integration[bot] in #44844
test(integration): hold the wire barrier until the test releases it or the wire closes by @devin-ai-integration[bot] in #45124
fix(vertex-ai): correct claude-haiku-5-5 thinking and forced tool flags by @berriai-litellm-provider-info-sync[bot] in #45116
feat(helm): allow custom labels, annotations, command and args on migrationJob by @ankitasahni511 in #42242
test(e2e): tag a2a, access_control, other, secret_manager and migrations tests with Subject metadata by @ryan-crabbe-berri in #44965
fix(azure): correct azure_ai/claude-haiku-5-5 forced tool use and thinking flags by @berriai-litellm-provider-info-sync[bot] in #45114
perf(lens): claim worker jobs from an indexed due queue instead of scanning every lens by @devin-ai-integration[bot] in #45095
perf(lens): prune ClickHouse partitions when sampling and sample in one pass by @devin-ai-integration[bot] in #45087
fix(logging): bound data URI regex so base64 truncation stays linear by @devin-ai-integration[bot] in #45132
perf(lens): bound single trace reads by the sampled start time by @devin-ai-integration[bot] in #45088
feat(lens-ui): show findings ranked by priority with frequency and highlighted evidence by @ishaan-berri in #45143
fix(ollama): answer in text after a tool result and cover ollama e2e on all three endpoints by @devin-ai-integration[bot] in #45079
fix(openai-compat): send provider attribution headers on the default SDK path (+ Perplexity) by @ryan-crabbe-berri in #44291
ci(codeql): run the default suite on full scans and security-extended on PRs by @devin-ai-integration[bot] in #45149
feat(lens): flag traces with global System 1 signals by @devin-ai-integration[bot] in #45094
feat(ui): configure Anthropic workload identity federation from the dashboard by @devin-ai-integration[bot] in #44889
feat(guardrails): add logging_only_scope to observe input, output, or both by @devin-ai-integration[bot] in #43695
test: move the unit half of 126 mixed legacy files into tests/unit by @devin-ai-integration[bot] in #45090
fix(model_prices): consolidate claude-haiku-5-5 over-100k pricing and capability flags by @devin-ai-integration[bot] in #45151
feat(ui): add auto-router usage and savings table by @tin-berri in #45152
refactor(rust-bridge): unify native call inputs and Messages settings by @yujonglee-berri in #45126
test(e2e): move the harness self-tests out of tests/e2e by @ryan-crabbe-berri in #45172
test(e2e): record steps for raw transport calls and poll helpers by @ryan-crabbe-berri in #45150
feat(lens): link traces to the conversation that started them by @ishaan-berri in #45169
fix(router): resume sync streaming fallbacks without retrying primary by @amarrtech in #43959
feat(guardrails): extend Akto guardrail to responses, MCP tools, attachments and masking by @rohan-akto in #44343
fix(streaming): stop re-wrapping a bridged stream's MidStreamFallbackError by @devin-ai-integration[bot] in #44989
fix(proxy): evict cached user on every proxy for tpm/rpm updates and edit limits in the users UI by @devin-ai-integration[bot] in #44130
perf(lens): classify signals within seconds of a trace finishing by @ishaan-berri in #45186
feat(proxy): add denied_passthrough_routes deny list for custom pass-through endpoints by @devin-ai-integration[bot] in #44924
fix(responses): keep prompt_cache_breakpoint markers in the chat to responses bridge by @devin-ai-integration[bot] in #44119
fix(mcp): preserve client application type during registration by @joshua-berri in #45159
feat(mcp): rate limit all MCP operations and add server-level rpm by @devin-ai-integration[bot] in #44600
fix(ci): align misc unit tests with NativeCall bridge and widened e2e diff gates by @devin-ai-integration[bot] in #45180
fix(types): serialize deferred pydantic schema builds across threads by @devin-ai-integration[bot] in #45034
fix(anthropic): normalize images for provider token counting by @tin-berri in #45185
feat(lens): scope traces to one agent with a header picker by @ishaan-berri in #45202
chore(codeowners): replace kerry-berri with kerrylu-berri by @devin-ai-integration[bot] in #45173
test: delete 73 legacy tests covered by e2e, unable to fail, or dead in CI by @devin-ai-integration[bot] in #45195
feat(lens): isolate ingestion and investigations in a Rust service by @moe-berri in #45148
fix(mcp): honor scoped cache freshness by @joshua-berri in #45165
chore(ui): bump next to 16.3.8 by @yuneng-berri in #45207
refactor(rust-bridge): share field and response marshaling by @yujonglee-berri in #45187
fix(lens): preserve numeric tags and report OTLP error codes by @moe-berri in #45206
fix(scheduler): remove a request's queue entry once it stops waiting by @RachelHuangZW in #43061
fix(vertex_ai): dial the multi-region Live API host for realtime sessions by @mateo-berri in #45166
feat(proxy,ui): add Moyai to the view switcher with quick connect by @ishaan-berri in #45196
fix(ci): restore vertex model sets mutated by get_optional_params tests by @devin-ai-integration[bot] in #45209
feat(decisions): add the OpenAI Decisions spec types and the System One translation by @devin-ai-integration[bot] in #45129
fix(anthropic): let /v1/messages mid-stream failures reach the proxy failure boundary by @D41910 in #44800
test: remove dead imports and helpers left behind by legacy test deletion by @devin-ai-integration[bot] in #45208
fix(vertex-ai): halve claude-sonnet-5-5 cache read price by @berriai-litellm-provider-info-sync[bot] in #45217
fix(anthropic): fail closed when a token-file or inline-token federation credential is missing a rule or organization id by @mateo-berri in #45178
fix(bedrock): map Scheduled batch jobs to in_progress on retrieve by @mrinal-berri in #45157
feat(ui): redesign usage page with stacked model and agent charts by @ishaan-berri in #45221
fix(bedrock): lower claude sonnet 5.5 cache read price to the aws offer index by @berriai-litellm-provider-info-sync[bot] in #45216
fix(responses): keep cache breakpoints on blocks the bridge stringifies by @mateo-berri in #40032
fix(bedrock_mantle): lower claude sonnet 5.5 cache read price to match bedrock runtime by @berriai-litellm-provider-info-sync[bot] in #45223
fix(ui): address usage redesign review findings by @ishaan-berri in #45229
perf(lens): faster trace opens and list pages at scale by @ishaan-berri in #45228
refactor(proxy): expose public names for private proxy helpers by @mateo-berri in #45170
fix(ollama): one stream id, 400 on non-text tool content, separated tool results, no empty message item by @mateo-berri in #45177
feat(lens): simplify deployment and first trace setup by @moe-berri in #45230
fix(proxy): traceparent/baggage fallback must not override caller metadata by @devin-ai-integration[bot] in #43688
fix(proxy): accept router-wide default_litellm_params in required-param validation by @yucheng-berri in #44483
fix(proxy): keep idle responses websockets open until a configurable session limit by @devin-ai-integration[bot] in #44433
fix(lens-ui): open traces on the top-level step and stop showing model names as input by @ishaan-berri in #45254
refactor(decisions): dispatch /v1/decisions through provider configs and the shared HTTP handler by @devin-ai-integration[bot] in #45130
feat(lens): show agent, user and slack thread first in the run header by @ishaan-berri in #45261
fix(proxy): declare the Moyai settings write's service target and allowlist its routes by @devin-ai-integration[bot] in #45220
feat(lens): show end-user feedback on traces, stored in ClickHouse by @ishaan-berri in #45171
fix(proxy): admit litellm_proxy/hosted_vllm in provider-endpoint discovery by @mayuriphad in #38617
feat(decisions): serve System One format at /v1/systemone and OpenAI format at /v1/decisions by @mateo-berri in #45184
feat(decisions): add OpenAI as a Decisions provider behind a shared decisions format by @mateo-berri in #45214
feat(lint): add LIT015 requiring pydantic models to be frozen by @devin-ai-integration[bot] in #42348
fix(router): keep include_fallback_errors off the provider call on the sync Router path by @mateo-berri in #45218
fix(router): honor a per-request fallbacks list on a mid-stream fallback by @mateo-berri in #45227
fix(responses): map input_audio blocks in the chat-to-Responses bridge by @mateo-berri in #45224
fix(proxy): answer a rejected websocket handshake without crashing the HTTP exception handler by @mateo-berri in #45251
ci(lint): gate every rule on its merge-base count and cap Anys at a fixed total by @devin-ai-integration[bot] in #40193
test(e2e): cover responses API gaps on gemini, azure, compact and context management by @devin-ai-integration[bot] in #45248
chore: bump litellm-proxy-extras to 0.4.107 and litellm-enterprise to 0.1.75 by @devin-ai-integration[bot] in #45291

New Contributors

@ankitasahni511 made their first contribution in #42242
@amarrtech made their first contribution in #43959
@rohan-akto made their first contribution in #44343
@RachelHuangZW made their first contribution in #43061
@D41910 made their first contribution in #44800
@mrinal-berri made their first contribution in #45157
@mayuriphad made their first contribution in #38617

Full Changelog: v1.106.0-dev.1...v1.106.0-dev.2