Repository navigation
Tags
Tags
release v1.30.0 [v1.30.0](v1.29.0...v1.30.0) (2026-11-03) Features: * xai: expose normalized watermark metadata (#1238) by mikehostetler * pass extra_metadata through to request telemetry events (#1223) by Amber-Cloud * minimax: forward reasoning effort and thinking controls (#1169) by Gilbert * bedrock: send structured output on the bedrock-mantle Messages API (#1167) by adriencohen * inference: add first-class Inference.net provider (#1154) by mikehostetler * cheaperinference: add Cheaper Inference provider (#1142) by aiapienthusiast * openai: accept context state with images in Decisions evaluations (#1147) by mhsdef * retry: let callers choose the HTTP statuses that requests retry (#1143) by Gilles Dubuc Bug Fixes: * google_vertex: handle non-2xx responses (#1237) by mikehostetler * deps: require patched JSV and unblock offline CI (#1235) by mikehostetler * mistral: accept low and medium reasoning effort (#1228) by mikehostetler * ignore invalid retry-after headers (#1224) by mikehostetler * preserve streaming error compatibility (#1219) by mikehostetler * test: restore environment variables changed by ProviderCase (#1198) by Tom Jaster * bedrock: route GPT-6 models on /openai/v1 and accept every service tier (#1166) by adriencohen * google: expose tool-use input tokens (#1165) by mikehostetler * stop sending message metadata in OpenAI-compatible requests (#1163) by Gustavo Honorato * google: accept only a boolean for google_url_context (#1161) by Peter Hoeg * opentelemetry: count cache reads and writes in input tokens (#1160) by Yannis Weishaupt * honor openai_structured_output_mode :tool_strict on Azure (#1159) by Arton Ragsdale * xai: drop stop and penalty options on reasoning models (#1158) by pcbeingused333 * google: preserve text thought signatures across continuations (#1127) by Gabriel Rinaldi * fireworks_ai: omit chat options from embeddings (#1152) by mikehostetler * timeout: preserve OpenTelemetry context in deadline tasks (#1149) by Gilles Dubuc * preserve message names in loose maps (#1146) by mikemikimike Refactoring: * route streaming through request runtime (#1218) by mikehostetler * route workflow requests through v1 runtime (#1216) by Tom Jaster * migrate buffered generation to v1 runtime (#1214) by Tom Jaster * migrate single-request operations to v1 runtime (#1215) by Tom Jaster * add v1 runtime and migrate OCR execution (#1213) by Tom Jaster * add internal v1 provider adapter (#1212) by Tom Jaster * generalize request plans for v1 operations (#1203) by Tom Jaster * separate pure stream decoding from StreamServer (#1205) by Tom Jaster * test: split OpenAI Responses API tests by responsibility (#1210) by Tom Jaster * test: split Azure provider tests by responsibility (#1209) by Tom Jaster * test: split Bedrock provider tests by responsibility (#1208) by Tom Jaster * test: split Anthropic provider tests by responsibility (#1207) by Tom Jaster * test: split Google provider tests by responsibility (#1206) by Tom Jaster * test: normalize the canonical v1.x test layout (#1202) by Tom Jaster * providers: remove redundant defaults macro uses (#1199) by Tom Jaster
release v1.28.0 [v1.28.0](v1.27.0...v1.28.0) (2026-10-09) Features: * add StepFun chat and audio providers (#1087) by mikehostetler * openai: honour image detail on Responses API input_image parts (#1084) by KLoevengreen Bug Fixes: * preserve Codex reasoning across tool turns (#1090) by mikehostetler * coverage: validate catalog community providers with live LLM API fixtures (#1089) by denys-rekun * preserve streamed cache TTLs and native OpenAI billing facts (#1081) by Matthew Neel * streaming: use concrete Mint HTTP2 type (#1083) by mikehostetler * openai_codex: decode terminal stream errors (#1082) by mikehostetler * deps: update Zoi and compatible Jido dependencies by mikehostetler
release v1.27.0 [v1.27.0](v1.26.0...v1.27.0) (2026-10-08) Features: * azure: support Chat Completions logprobs (#1073) by Arton Ragsdale * allow Gemini 3 object output with tools (#1077) by mvanhorn * openai: support Decisions evaluations (#1075) by mikehostetler Bug Fixes: * schema: bound validator cache memory (#1078) by mikehostetler * redact unknown provider reasoning payloads (#1070) by Wu Shuwen * close shared contract gaps from design review (#1069) by mikehostetler * openai: enforce anonymous auth and exact streaming model identity (#1065) by Justin Smestad * openai: decode Responses refusals as text with content_filter finish reason (#1064) by Nickjgniklu * http: prevent non-streaming connection pool timeouts (#1067) by mikehostetler
release v1.26.0 [v1.26.0](v1.25.0...v1.26.0) (2026-09-29) Features: * openai: add DevDay Responses and model support (#1063) by mikehostetler Bug Fixes: * openai: read nested error details from Responses stream error events (#1061) by Diaphel Thompson * bedrock: send output_config.effort for adaptive-thinking Claude models (#1055) by jhlee111 * streaming: honor per-request proxy connection options (#1054) by mikehostetler
release v1.25.0 [v1.25.0](v1.24.0...v1.25.0) (2026-09-25) Features: * Responses API reasoning summaries, reasoning.context and compaction for OpenAI and Azure (#1048) by fv-teacherspace * price conditional LLMDB tariffs from complete usage (#1046) by houllette * Image streaming support for OpenAI (#1043) by Vasilis Spilka * openai: carry assistant phase on each output text part (#1047) by Mark Ericksen * streaming: support in-process providers (#1040) by mikehostetler * Add transparent option for openai images (#1037) by Vasilis Spilka * forward OpenRouter evaluation provider options (#1038) by fschucht * router: add dynamic model router spike (#1036) by mikehostetler * streaming: carry provider response ids on streamed responses (#1023) by dankentfield-sona * generation: dispatch catalog OpenAI chat gateways from execution metadata (#1028) by mikehostetler * support OpenRouter Jev evaluation (#1027) by mikehostetler Bug Fixes: * anthropic: report Claude Code 2.1.282 on subscription OAuth requests (#1052) by neilberkman * preserve non-strict Responses tool schemas (#1049) by mikehostetler * deps: update Mint for security advisory by mikehostetler * usage: distinguish cache reads and writes (#1035) by mikehostetler * bedrock: surface mid-stream exceptions and map all Converse stop reasons (#1033) by adriencohen * bedrock: send serviceTier on Converse and honor explicit region (#1032) by adriencohen * anthropic: replay tool search blocks in place (#1034) by mikehostetler * deepseek: map reasoning effort levels to supported values (#1025) by Matthew Neel Performance: * context: batch message assembly (#1031) by mikehostetler
release v1.24.0 [v1.24.0](v1.23.0...v1.24.0) (2026-09-17) Features: * evaluation: add TypeSafe Jev evaluation API (#1022) by mikehostetler * lmstudio: add LM Studio provider (#1020) by Michael Mazurczak Bug Fixes: * openai: keep replayed reasoning items at their own turn in Responses input (#1016) by Michael Mazurczak
release v1.23.0 [v1.23.0](v1.22.0...v1.23.0) (2026-09-15) Features: * add tool search for Anthropic, Bedrock, and OpenAI Responses (#1013) by mikehostetler * bedrock: support prompt caching on the Converse API (#1006) by adriencohen * bedrock: return document citations on the Converse API (#1005) by adriencohen * bedrock: send documents and video on the Converse API (#1004) by adriencohen * codex: support caller-owned turn attribution (#1003) by Danila Poyarkov * bedrock: guard selected content parts on the Converse API (#1002) by adriencohen * google: support validated function calling mode (#1001) by Javier Cuevas * bedrock: round-trip reasoning on the Converse API (#997) by adriencohen * bedrock: support guardrails (#995) by adriencohen * openai: keep a builtin tool call's status and mark failed calls as error spans (#991) by Vasilis Spilka Bug Fixes: * bedrock: route Mantle models through the correct API (#1014) by mikehostetler * model: resolve prefixes consistently across spec formats (#1012) by mikehostetler * normalize numeric-string usage counters (#1009) by mikehostetler * codex: preserve session and prompt cache identity across transports (#996) by Danila Poyarkov * deps: update ex_doc and vulnerable igniter (#994) by dependabot[bot] * azure: accept include in provider_options for Responses API models (#990) by Vasilis Spilka
release v1.22.0 [v1.22.0](v1.21.1...v1.22.0) (2026-09-04) Features: * openai: add experimental GPT-6 Astra Responses support (#987) by mikehostetler * bedrock: serve requests through the bedrock-mantle endpoint (#979) by adriencohen * bedrock: support application inference profiles (#978) by adriencohen Bug Fixes: * deps: update vulnerable example dependencies by mikehostetler * deps: allow monthly llm_db catalog updates (#988) by BlueHotDog * embedding: forward dimensions to compatible providers (#986) by mikehostetler * zai: include tool_call_id and name in tool result messages (#985) by Bill Huang
release v1.21.1 [v1.21.1](v1.21.0...v1.21.1) (2026-08-26) Bug Fixes: * openai: decode response.failed and error stream events as terminal error chunks (#977) by Eric Draut * anthropic: read thinking tokens from output_tokens_details (#976) by brent-emb * tool: ignore unknown schema input keys (#975) by Niklas Johansson * google: respect model thinking levels (#972) by mikehostetler
PreviousNext