# Changelog
All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
## [Unreleased]
<!-- changelog -->
## [v1.9.0](https://github.com/agentjido/req_llm/compare/v1.8.0...v1.9.0) (2026-03-27)
### Features:
* add mistral provider (#554) by mikehostetler
* add `thinkingLevel` support to Google provider (#565) by Tom Taylor
* add `thinkingLevel` support to Google provider by Tom Taylor
* add `finch_request_adapter` config to modify streaming requests (#566) by johantell
* add FinchRequestAdapter for config-level streaming request transforms by johantell
* add OpenAI websocket streaming and realtime support (#559) by mikehostetler
* allow unknown models with warnings instead of errors (#561) by dl-alexandre
* add reranking API with batch processing (#553) by mikehostetler
* add OpenTelemetry bridge and telemetry stubs (#548) by mikehostetler
* add video URL message support (#549) by mikehostetler
* support Anthropic native tool options (#547) by mikehostetler
* support Anthropic native tool options by mikehostetler
* add Google Imagen image generation support (#550) by mikehostetler
* generation: repair lightly malformed structured JSON (#543) by mikehostetler
* add available model discovery (#544) by mikehostetler
* add application-layer cache hooks (#541) by mikehostetler
* generation: add application cache hooks by mikehostetler
* rate limit retry (429) and custom headers for streaming (#537) by dl-alexandre
* pass custom headers to Finch streaming requests by dl-alexandre
* Add Deepseek support (#538) by Luis Ezcurdia
### Bug Fixes:
* openai: retrieve structured output from tool call (#572) by Robin Verton
* openai: retrieve structured output from tool call by Robin Verton
* openai: validate tool-call structured output by Robin Verton
* track google thinking levels in telemetry by Tom Taylor
* restore streaming fixture recording functionality (#560) by dl-alexandre
* add Google long-context pricing tiers (#546) by mikehostetler
* Allow model specific base_url for streaming responses (#562) by meanderingstream
* Support json schema usage for gemini 3.1 (#563) by Akash Khan
* preserve Anthropic tool formatter shape by mikehostetler
* default GPT-4o models to OpenAI Responses API (#552) by mikehostetler
* zero app-cache usage on response hits (#542) by mikehostetler
* usage: zero app cache hit usage by mikehostetler
* remove dead cache provider meta fallback by mikehostetler
* support multiple system messages (#545) by mikehostetler
* register azure request options from defaults by mikehostetler
* schema: preserve doc option in nested {:map, opts} fields by dl-alexandre
* openai: validate nested object schemas with string keys by dl-alexandre
* streaming retry accumulator and 429 detection bugs by dl-alexandre
* streaming: retry 429 responses and honor map headers by dl-alexandre
* streaming: resolve 429 retry dialyzer warning by dl-alexandre
* streaming: restore 429 failure callback order by dl-alexandre
* bedrock: drop empty messages in Converse encoder (#540) by Julian Scheid
### Refactoring:
* simplify transient streaming logs by mikehostetler
## [v1.8.0](https://github.com/agentjido/req_llm/compare/v1.7.1...v1.8.0) (2026-03-23)
### Features:
* add web_fetch server tool support by Tonyhaenn
* add oauth-backed OpenAI Codex routing (#513) by l3wi
* openai: add oauth-backed codex routing by l3wi
* add native request and reasoning telemetry (#515) by mikehostetler
* add native request and reasoning telemetry by mikehostetler
* add cross-provider tool call ID compatibility (#417) by mikehostetler
* add cross-provider tool call id compatibility by mikehostetler
### Bug Fixes:
* honor req_llm load_dotenv when starting llm_db (#536) by mikehostetler
* honor req_llm load_dotenv when starting llm_db by mikehostetler
* include llm_db in dialyzer plt apps by mikehostetler
* openai: validate nested object schemas with string keys (#533) by dl-alexandre
* bedrock: filter empty text blocks in Converse (#534) by Julian Scheid
* schema: preserve doc option in nested {:map, opts} fields (#532) by dl-alexandre
* retry transient streaming transport failures (#531) by mikehostetler
* preserve google embedding usage metadata (#530) by mikehostetler
* break azure anthropic xref cycle (#529) by mikehostetler
* make livebook validation deterministic by dl-alexandre
* format livebook parse errors from metadata by dl-alexandre
* remove unsupported outputMimeType and add response_modalities fo⦠(#517) by mrdotb
* remove unsupported outputMimeType and add response_modalities for Google images by mrdotb
* normalize Gemini response modalities by mrdotb
* gate Anthropic server tools behind beta headers by Tonyhaenn
* default Codex streaming auth to oauth by l3wi
* merge consecutive same-role entries in Gemini message encoding (#525) by paulorumor
* prevent Inspect.ReqLLM.Context crash on Logger depth truncation (#519) by Edgar Gomes
* azure: add Grok to known model families (#518) by shelvick
* azure: add Grok to known model families by shelvick
* azure: add Grok to OpenAI-compatible prefix lists and add reasoning_token_budget by shelvick
* support image content in Bedrock tool results (#516) by Julian Scheid
* OpenAI content encoding for thinking parts and empty arrays (#428) by laudney
* restore telemetry raw payload capture (#520) by mikehostetler
* restore telemetry raw payload capture by mikehostetler
* resolve CI regressions for telemetry by mikehostetler
* harden telemetry reasoning normalization by mikehostetler
* update tool call IDs and timestamps in compatibility JSON fixtures by mikehostetler
* accept map-based vertex model metadata by mikehostetler
* google: strip atom-keyed forbidden schema fields (#504) by pcharbon70
* remove impossible role == :tool checks (#508) by Barna Kovacs
* remove redundant fallback for reasoning tokens in normalize (#507) by Barna Kovacs
### Refactoring:
* only inherit llm_db load_dotenv when unset by mikehostetler
## [v1.7.1](https://github.com/agentjido/req_llm/compare/v1.7.0...v1.7.1) (2026-03-14)
### Bug Fixes:
* make image tests resilient to llm_db catalog updates by mikehostetler
* preserve assistant metadata and reasoning details when normalizing loose maps (#501) by Chris Lema
* format context normalization branches by Chris Lema
## [v1.7.0](https://github.com/agentjido/req_llm/compare/v1.6.0...v1.7.0) (2026-03-14)
### Features:
* normalize and document inline model specs (#500) by mikehostetler
* normalize and document inline model specs by mikehostetler
* add Alibaba Cloud Bailian (DashScope) provider (#470) by Steven Holdsworth
* auth: support OAuth tokens and codex Responses routing (#478) by l3wi
* auth: support OAuth tokens and codex Responses routing by l3wi
* Uses the same Finch Pool for streaming/non-streaming (#466) by ycastorium
* Add Usage to Embeddings (#444) by ycastorium
* add response_format to Vertex AI provider schema for OpenAI-compat MaaS models (#450) by shelvick
* pass anthropic_beta through Bedrock to request body (#468) by stevehodgkiss
* add Igniter installer for automated package setup by Nickcom4
### Bug Fixes:
* remove unreachable inline model clause by mikehostetler
* honor Google image output format (#499) by mikehostetler
* emit usage for Vertex embeddings (#498) by mikehostetler
* GoogleVertex extract_gcp_credentials from provider_options (#496) by paulorumor
* streaming: propagate transport errors instead of swallowing (#491) by Julian Scheid
* streaming: propagate transport errors instead of swallowing by Julian Scheid
* move examples into nested mix project by Dave Lucia
* docs: clean up sidebar (#490) by Dave Lucia
* docs: clean up sidebar by Dave Lucia
* docs: restore public providers and usage docs by Dave Lucia
* docs: improve exdoc module grouping by Dave Lucia
* ReqLLM.Response.object returns string keys (#488) by Dave Lucia
* expose stream keepalive callbacks (#495) by mikehostetler
* clean up StreamServer processes after streaming completes (#494) by mikehostetler
* clean up stream servers after stream completion by mikehostetler
* harden stream server lifecycle cleanup by mikehostetler
* auth: honor auth_mode and string-key wire protocol by l3wi
* Fixes sse in case of server failure (#463) by ycastorium
* Fixes sse in case of server failure by ycastorium
* Adds a new termination field by ycastorium
* preserve stop finish_reason for buffered done event by ycastorium
* exclude embedding return_usage from provider transport options (#471) by mikehostetler
* normalize nested string-key tool arguments safely (#459) by mikehostetler
* tool: normalize nested string-key tool arguments by mikehostetler
* tool: normalize map keys in union and tuple types by mikehostetler
* core: remove unsafe runtime string-to-atom conversions by mikehostetler
* ReqLLM embed should use embedding schema, not generation (#446) by ycastorium
* handle API error responses in Vertex AI OpenAI-compat endpoint (#451) by shelvick
* handle API error responses in Vertex AI OpenAI-compat parse_response by shelvick
* unwrap single-element list body in OpenAICompat.parse_response by shelvick
* bedrock: update inference profile prefixes (#449) by stevehodgkiss
* stream usage normalization and accumulation (#464) by stevehodgkiss
* encode tool outputs inline in Responses API input array (#454) by austin macciola
* only add cache_control to last tool in Anthropic prompt caching (#453) by tomtrin
* remove $schema and additionalProperties recursively in to_google_format (#460) by Jhon Pedroza
* allow max_retries to be 0 to disable retries (#469) by stevehodgkiss
* omit mimeType from fileData for YouTube and extensionless URLs (#442) by Nickcom4
* omit mimeType from fileData for YouTube and extensionless URLs by Nickcom4
* google: return nil for unknown inferred mime type by Nickcom4
* strip redundant tool_choice auto from OpenRouter requests (#437) by Noah S
* strip redundant tool_choice auto from OpenRouter requests by Noah S
* widen generation function specs to match Context.normalize/2 (#441) by Nickcom4
* widen generation function specs to match Context.normalize/2 by Nickcom4
* Removes redundant validation (#448) by ycastorium
## [v1.6.0](https://github.com/agentjido/req_llm/compare/v1.5.1...v1.6.0) (2026-02-19)
### Features:
* implement #431 response classify, usage API, and tool key safety (#432) by mikehostetler
* add response classify API and harden usage/tool normalization by mikehostetler
* add igniter installer (#410) by AdwayKasture
* add embedding support to google_vertex provider (#423) by paulorumor
* enable reasoning support for DeepSeek models on Azure (#412) by shelvick
* enable reasoning support for DeepSeek models on Azure by shelvick
* Add on_tool_call callback to StreamResponse (#413) by Arjan Scherpenisse
* add OpenAI-compatible model family support for Vertex AI MaaS models (#422) by shelvick
* jido: enhance agent functionality with usage tracking and multi-turn support by mikehostetler
### Bug Fixes:
* Updating model data (#435) by Pedro Assunção
* resolve dialyzer warnings in response classify normalization by mikehostetler
* handle string finish reasons in response classify by mikehostetler
* add Gemini 3 thought_signature support for function calls (#427) by Brandon L'Europa
* Do not force strict tools for responses api (#399) by ycastorium
* use proportional estimate for inferred reasoning tokens by shelvick
* route Vertex AI MaaS models to endpoints/openapi/chat/completions (#424) by shelvick
* increase default timeout for DeepSeek and MAI-DS models on Azure (#425) by shelvick
* Azure AI Foundry auth header and API version defaults (#411) by shelvick
* remove invalid `id` field from Google functionCall serialization (#414) by paulorumor
* remove invalid `id` field from Google functionCall serialization by paulorumor
* pass strict flag through to Anthropic tool format (#415) by Edgar Gomes
* pass strict flag through to Anthropic tool format by Edgar Gomes
* pass strict flag through to Bedrock Converse tool format by Edgar Gomes
* harden ZAI thinking re-encoding and req_llm.gen model errors (#420) by mikehostetler
* harden zai thinking re-encoding and gen model validation by mikehostetler
* zai: remove unreachable encode_zai_content clauses by mikehostetler
* streaming decode for inference profile models using InvokeModel API (#406) by neilberkman
* req_llm: improve streaming stability with finch client fixes by mikehostetler
## [v1.5.1](https://github.com/agentjido/req_llm/compare/v1.5.0...v1.5.1) (2026-02-04)
### Bug Fixes:
* Return metadata for incomplete responses (#403) by Tom Duffield
* include metadata with incomplete responses by Tom Duffield
* add missing verbosity documention for openai by Tom Duffield
* improve streaming stability with finch client fixes (#400) by mikehostetler
* req_llm: improve streaming stability with finch client fixes by mikehostetler
* move changelog marker so git_ops inserts in correct position by mikehostetler
### Refactoring:
* Centralize text extraction and update dependencies by mikehostetler
## [v1.5.0](https://github.com/agentjido/req_llm/compare/v1.4.1...v1.5.0) (2026-02-01)
### Features:
* add xAI image generation support (#397) by Victor
* add xai image generation support by Victor
### Bug Fixes:
* remove unsupported-option warnings by Victor
## [1.4.1] - 2026-01-31
### Added
- Tool call normalization helpers: `ToolCall.from_map/1` and `ToolCall.to_map/1` for consistent tool-call handling across providers (#396)
### Fixed
- Made `git_ops` configuration available outside dev-only config so CI releases work correctly
## [1.4.0] - 2026-01-30
### Added
- Comprehensive usage and billing infrastructure with richer usage/cost reporting (#371)
- Reasoning cost breakdown with `reasoning_cost` field in cost calculations (#394)
- OpenRouter enhancements:
- `openrouter_usage` and `openrouter_plugins` provider options (#393)
- Native JSON schema structured output support (#374)
- Google provider options:
- `google_url_context` for URL grounding (#392)
- `google_auth_header` option for streaming requests (#382)
- OpenAI improvements:
- Configurable strict mode for JSON schema validation (#368)
- Verbosity support for reasoning models (#354)
- Cohere Embeddings on Bedrock (#365)
- Structured and multimodal tool outputs (#357)
- Model `base_url` override in model configuration (#366)
### Changed
- Replaced TypedStruct with Zoi schemas for data structures (#376)
### Fixed
- Image-only attachments validation for OpenAI and xAI (#389)
- `translate_options` changes now preserved in `provider_options` (#381)
- StreamServer termination handled gracefully in FinchClient (#379)
- Anthropic schema constraints stripped when unsupported (#378)
- `api_key` added to internal keys preventing leakage (#355)
## [1.3.0] - 2026-01-21
### Added
- **New Providers:**
- Zenmux provider and playground (#342)
- vLLM provider for self-hosted OpenAI-compatible models (#202)
- Venice AI provider (#200)
- Azure DeepSeek model support (#254)
- Azure Foundry Bearer token authentication (#338)
- Z.ai thinking parameter support (#303)
- OpenAI `service_tier` option (#321)
- OpenAI wire protocol routing (#318)
- Context and streaming improvements:
- `Context.normalize/1` extended for `tool_calls` and tool result messages (#313)
- Preserve `reasoning_details` during streaming tool-call round-trips (#300)
- `StreamResponse.classify/1` and `Response.Stream.summarize/1` (#311)
- Google file URI support for `image_url` content parts (#339)
- Reasoning signatures retainment (#344)
- `generate_object` now accepts map input (#301)
- OpenRouter support for google/gemini-3-flash-preview (#298)
### Fixed
- Anthropic `encrypted?` flag in reasoning details extraction
- Anthropic cache token handling for API semantics (#316)
- Missing reasoning levels (#332)
- Google Gemini thinking tokens in cost calculation (#336)
- Hyphenated tool names for MCP server compatibility (#323)
- Azure `provider_options` validation and ResponsesAPI `finish_reason` parsing (#266)
- Cache token extraction and cost calculation (#309)
- JSON arrays for JsonSchema and Gemini 3 schema calls (#310)
- Gemini `generate_object` always sets `responseMimeType` (#299)
- Z.ai `zai_coding_plan` provider support (#347)
- Ecosystem conflicts with typedstruct naming (#315)
## [1.2.0] - 2025-12-22
### Added
- Image generation support (#293)
- Anthropic web search support for models (#292)
- OpenRouter first-class `reasoning_details` support (#267)
- Google Vertex AI:
- Inline JSON credentials support (#260)
- Google Search grounding for Gemini models (#284)
- Anthropic message caching for conversation prefixes (#281)
- `load_dotenv` config option to control .env file loading (#287)
### Changed
- Response assembly unified across providers (#274)
- Streaming preserves grounding metadata (#278)
### Fixed
- Streaming errors propagate to `process_stream` result (#286)
- Debug URLs sanitized when streaming with Google (#279)
- Functional tool streaming response bug (#263)
## [1.1.0] - 2025-12-21
### Added
- **New Providers:**
- Azure OpenAI provider (#245)
- Google Vertex Gemini support
- Google Vertex AI Anthropic provider (#217)
- OAuth2 token caching for Google Vertex AI (#174)
- Google Context Caching for Gemini models (#193)
- Amazon Bedrock `service_tier` support (#225)
- OpenAI / Responses API:
- `tool_choice: required` support (#215)
- Reasoning effort support (#244)
- Model capability helper functions (#222)
- `StreamResponse.process_stream/2` for real-time callbacks (#178)
- Custom providers defined outside req_llm (#201)
- llm_db integration for model metadata (#212)
- Credential fallback for fixture recording (#218)
### Changed
- Data structures migrated to typedstruct (#256)
- Streaming metadata access made reusable (#206)
- Anthropic structured output modes enhanced (#223)
### Fixed
- Default timeout increased for OpenAI reasoning models (#252)
- Consecutive tool results merged into single user message (#250)
- `.env` loading respects existing env vars (#249)
- Responses API tool encoding uses flat structure (#247)
- `finish_reason` captured correctly when streaming (#241)
- OpenAI Responses context replay and Anthropic structured output decode (#228)
- StreamResponse context merging (#224)
- `tool_budget_for` pattern match regression from LLMDB integration (#221)
- `reasoning_overlay` pattern match for llmdb structure (#219)
- Missing `api_key` in Anthropic extra options (#216)
- Typespec for object generation to allow Zoi schemas (#208)
- Cerebras strict mode handling (#180)
- JSV schema validation preserves original data types (#173)
- Cached token extraction from Google API responses (#192)
## [1.0.0] - 2025-11-02
First production-ready release of ReqLLM.
### Added
- **Google Vertex AI provider** with comprehensive Claude 4.x support
- OAuth2 authentication with service accounts
- Full Claude model support (Haiku 4.5, Sonnet 4.5, Opus 4.1)
- Extended thinking and prompt caching capabilities
- **AWS Bedrock inference profile models** with complete fixture coverage
- Anthropic Claude inference profiles
- OpenAI OSS models
- Meta Llama inference profiles
- Cohere Command R models
- **Provider base URL override** capability via application config
- AWS Bedrock API key authentication (introduced by AWS in July 2025)
- Context tools persistence for AWS Bedrock multi-turn conversations
- Schema map-subtyped list support for complex nested structures
### Changed
- Google provider uses v1beta API as default version
- Streaming protocol callback renamed from `decode_sse_event` to `decode_stream_event`
### Fixed
- Groq UTF-8 boundary handling in streaming responses
- Schema boolean encoding preventing invalid string coercion
- AWS Bedrock Anthropic inference profile model ID preservation
- AWS Bedrock Converse API usage field parsing
- AWS Bedrock model ID normalization for metadata lookup
[Unreleased]: https://github.com/agentjido/req_llm/compare/v1.4.1...HEAD
[1.4.1]: https://github.com/agentjido/req_llm/compare/v1.4.0...v1.4.1
[1.4.0]: https://github.com/agentjido/req_llm/compare/v1.3.0...v1.4.0
[1.3.0]: https://github.com/agentjido/req_llm/compare/v1.2.0...v1.3.0
[1.2.0]: https://github.com/agentjido/req_llm/compare/v1.1.0...v1.2.0
[1.1.0]: https://github.com/agentjido/req_llm/compare/v1.0.0...v1.1.0
[1.0.0]: https://github.com/agentjido/req_llm/releases/tag/v1.0.0
