What's Changed
- streaming gate fixes by @akshaydeo in #7331
- stream gate memory fixes by @akshaydeo in #7332
- adds more gating test cases by @akshaydeo in #7333
- enterprise docs by @akshaydeo in #7334
- enterprise docs by @akshaydeo in #7335
- feat: extract session stickiness into a pluggable
SessionAffinityinterface with scoped, identity-aware key bindings by @Pratham-Mishra04 in #7176 - feat: route and key session affinity driven by observed outcomes, with
x-bf-session-affinityheader opt-out by @Pratham-Mishra04 in #7177 - feat: add
BifrostContextKeySessionAffinityfor per-request session affinity control and document provider-level session binding behavior by @Pratham-Mishra04 in #7178 - refactor: export
SessionStateKey,SessionAffinity, and related helpers; unify complexity session keying under shared session state by @Pratham-Mishra04 in #7179 - docs: add Session Affinity guide and cross-link it from routing, CLI agent, and request-options docs by @Pratham-Mishra04 in #7210
- fix: record session affinity agreement with routing in trail and engines-used, log unweighted provider exclusions, clarify docs by @Pratham-Mishra04 in #7289
- [fix]: preserve API web search source names by @g-yixuan in #7358
- fix(anthropic): carry allowed_tools and parallel_tool_calls into the request by @basil-k-aji-dev in #7348
- fix(bedrock): keep cache_control when a message's content is a string by @basil-k-aji-dev in #7360
- dependabot cleanups by @akshaydeo in #7378
- fix: allow all providers access and api models response by @TejasGhatte in #7375
- typesafe models support, new decisions endpoint by @akshaydeo in #7355
- feat(decision): emulate decisions via LLM tool-calling for all providers by @akshaydeo in #7361
- benchmarking and api struct fixes by @akshaydeo in #7384
- fix: removes unnecessary allocations from compat plugin by @sammaji in #7097
- fix: strip bedrock cachepoint markers for non-bedrock providers while preserving them for fallbacks by @sammaji in #7182
- adds bedrock unknown json document handling by @akshaydeo in #7396
- [fix]: preserve Grok 4.7 xhigh reasoning effort by @nettee in #7403
- docs: add prompt deployment docs by @roroghost17 in #7420
- adds safeguard support for passthrough requests by @akshaydeo in #7393
- memory fixes for streaming cancellation by @akshaydeo in #7406
- fix(responses): stop rewriting OpenAI computer tool to computer_use_preview by @abhishekgahlot2 in #7426
- perf: avoid redundant JSON decoding in Responses requests by @G-XD in #7412
- fix(schemas): deep copy extended Responses fields in DeepCopyResponsesMessage by @Madhuvod in #7422
- ci: add CodSpeed performance benchmarks for the request hot path by @codspeed[bot] in #7409
- readact anthropic billing headers to avoid cache spoiling by @akshaydeo in #7431
- kimi and deepseek support for claude code by @akshaydeo in #7430
- fix: claude opus 5-5 changes by @TejasGhatte in #7433
- fix: restore
HTTP_PROXY/HTTPS_PROXYsupport for MCP connections while blocking link-local targets via pre-proxy destination guard by @Pratham-Mishra04 in #7437 - test fixes by @akshaydeo in #7440
- fix: computer toolset compatibility by @TejasGhatte in #7434
- fix: computer toolset model capability by @TejasGhatte in #7441
- v2.2.2 by @akshaydeo in #7442
- changelogs for 2.2.2 by @akshaydeo in #7443
- adds preamble support for all openai model family by @akshaydeo in #7444
- preamble support for openai model family across the providers by @akshaydeo in #7445
- docs fixes by @akshaydeo in #7447
New Contributors
- @g-yixuan made their first contribution in #7358
- @basil-k-aji-dev made their first contribution in #7348
- @abhishekgahlot2 made their first contribution in #7426
- @codspeed[bot] made their first contribution in #7409
Full Changelog: ent-v2.2.1-base...ent-v2.2.2-base