<!--
SPDX-FileCopyrightText: 2024 ash_ai contributors <https://github.com/ash-project/ash_ai/graphs/contributors>
SPDX-License-Identifier: MIT
-->
# Change Log
All notable changes to this project will be documented in this file.
See [Conventional Commits](Https://conventionalcommits.org) for commit guidelines.
<!-- changelog -->
## [v0.6.1](https://github.com/ash-project/ash_ai/compare/v0.6.0...v0.6.1) (2026-04-13)
### Bug Fixes:
* use context in Ash reads in AshAI.Gen.Chat by [@zachdaniel](https://github.com/zachdaniel)
## [v0.6.0](https://github.com/ash-project/ash_ai/compare/v0.5.0...v0.6.0) (2026-04-09)
### Features:
* Hard cutover to ReqLLM: remove LangChain runtime and migrate chat generator (#177) by mikehostetler [(#177)](https://github.com/ash-project/ash_ai/pull/177)
* add mcp_ui_resource for MCP Apps spec (#173) by [@nallwhy](https://github.com/nallwhy) [(#173)](https://github.com/ash-project/ash_ai/pull/173)
### Improvements:
* For `gen.chat` task make all tools available by default (#176) by Kenneth Kostrešević [(#176)](https://github.com/ash-project/ash_ai/pull/176)
* Support --provider, --route and --live-component (#175) by Kenneth Kostrešević [(#175)](https://github.com/ash-project/ash_ai/pull/175)
* add `strict?` option, change how filters behave under it by [@zachdaniel](https://github.com/zachdaniel)
## [Unreleased]
### Breaking Changes
* Hard cutover from LangChain to ReqLLM for all LLM access paths.
* Removed `AshAi.setup_ash_ai/2`, `AshAi.functions/1`, and LangChain-based `AshAi.iex_chat/2`.
* Added ReqLLM-first orchestration APIs:
* `AshAi.list_tools/1`
* `AshAi.build_tools_and_registry/1`
* `AshAi.ToolLoop.run/2`
* `AshAi.ToolLoop.stream/2`
* `AshAi.iex_chat/1`
* `prompt/2` now uses ReqLLM model specs (`"provider:model"`, ReqLLM tuples, or functions returning them).
* Prompt actions keep `tools:` support via `AshAi.ToolLoop`.
* `mix ash_ai.gen.chat` now generates ReqLLM-based chat code and runtime config (`config :req_llm, ...`).
* Runtime does not include Jido integration.
### Migration Notes
* Full guide: [LangChain to ReqLLM Migration Guide](/documentation/topics/langchain-to-reqllm-migration.md)
* Replace LangChain model structs with ReqLLM model specs, e.g.:
* `LangChain.ChatModels.ChatOpenAI.new!(%{model: "gpt-4o"})`
* becomes `"openai:gpt-4o"`
* Replace `AshAi.setup_ash_ai/2` usage with `AshAi.ToolLoop` or `AshAi.build_tools_and_registry/1`.
* Update runtime configuration keys:
* `config :langchain, ...` -> `config :req_llm, ...`
### Improvements
* prompt actions: default tool-loop `max_iterations` to `:infinity` (unless explicitly set)
* prompt actions: return tool-loop failures as action errors with reason details
* prompt actions: support `verbose?` tool-loop lifecycle logging
* prompt actions: use permissive schema for unconstrained `:map` returns
* `ash_ai.gen.chat`: preserve and preview tool call arguments in both LiveView and LiveComponent templates
* `ash_ai.gen.chat`: collapse generated tool UI helpers into `AshAi.ChatUI.Tools.extract/1` with one-time warning flash on malformed tool data
* docs: add legacy compatibility notes for adapter mapping and embedding return tuple behavior
## [v0.5.0](https://github.com/ash-project/ash_ai/compare/v0.4.0...v0.5.0) (2026-01-26)
### Features:
* Add ChatOpenAiResponses support (#163) by sezaru [(#163)](https://github.com/ash-project/ash_ai/pull/163)
* Add ChatOpenAiResponses support by sezaru [(#163)](https://github.com/ash-project/ash_ai/pull/163)
* Use StructuredOutput for ChatOpenAiResponses by sezaru [(#163)](https://github.com/ash-project/ash_ai/pull/163)
* Add support for modifying the chain in prompt (#142) by [@Munksgaard](https://github.com/Munksgaard) [(#142)](https://github.com/ash-project/ash_ai/pull/142)
### Bug Fixes:
* add additionalProperties constraint to input schema (#144) by [@kingdomcoding](https://github.com/kingdomcoding) [(#144)](https://github.com/ash-project/ash_ai/pull/144)
* fix issue with AshOban compiling before AshAi action definition (#148) by Aaron Westbrook [(#148)](https://github.com/ash-project/ash_ai/pull/148)
### Improvements:
* Handle anonymous function in tool load definition (#157) by josh-brainbox [(#157)](https://github.com/ash-project/ash_ai/pull/157)
* add arguments to tools by josh-brainbox [(#157)](https://github.com/ash-project/ash_ai/pull/157)
## [v0.4.0](https://github.com/ash-project/ash_ai/compare/v0.3.0...v0.4.0) (2025-11-23)
### Features:
* Added _meta field to tools MCP (#140) by [@C-Sinclair](https://github.com/C-Sinclair) [(#140)](https://github.com/ash-project/ash_ai/pull/140)
* Add MCP resources support with action descriptions (#138) by [@C-Sinclair](https://github.com/C-Sinclair) [(#138)](https://github.com/ash-project/ash_ai/pull/138)
## [v0.3.0](https://github.com/ash-project/ash_ai/compare/v0.2.14...v0.3.0) (2025-10-28)
### Features:
* langchain 0.4 (#132) by [@barnabasJ](https://github.com/barnabasJ) [(#132)](https://github.com/ash-project/ash_ai/pull/132)
### Bug Fixes:
* turn content into string for chat title by [@barnabasJ](https://github.com/barnabasJ) [(#132)](https://github.com/ash-project/ash_ai/pull/132)
### Improvements:
* update ash_oban to 0.5 by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.14](https://github.com/ash-project/ash_ai/compare/v0.2.13...v0.2.14) (2025-10-16)
### Improvements:
* allow multiple full text definitions in vectorize section (#125) by Abdessabour Moutik [(#125)](https://github.com/ash-project/ash_ai/pull/125)
## [v0.2.13](https://github.com/ash-project/ash_ai/compare/v0.2.12...v0.2.13) (2025-09-27)
### Bug Fixes:
* minor QOL improvement to the redirection to other pages (#120) by Abdessabour Moutik [(#120)](https://github.com/ash-project/ash_ai/pull/120)
* ash_ai.gen.chat to validate text presence in messages (#119) by Daniel Hoelzgen [(#119)](https://github.com/ash-project/ash_ai/pull/119)
* BadMapError when LangChain/MCP calls tools without arguments (#118) by [@matthewsinclair](https://github.com/matthewsinclair) [(#118)](https://github.com/ash-project/ash_ai/pull/118)
### Improvements:
* don't install usage rules as part of installing ash ai by [@zachdaniel](https://github.com/zachdaniel)
* Support LangChain 0.4 (#124) by Arjan Scherpenisse [(#124)](https://github.com/ash-project/ash_ai/pull/124)
## [v0.2.12](https://github.com/ash-project/ash_ai/compare/v0.2.11...v0.2.12) (2025-08-31)
### Bug Fixes:
* pass context option through setup_ash_ai to nested actions (#111) by [@bradleygolden](https://github.com/bradleygolden)
### Improvements:
* don't show input if no inputs to action by [@zachdaniel](https://github.com/zachdaniel)
* add `action_parameters` option by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.11](https://github.com/ash-project/ash_ai/compare/v0.2.10...v0.2.11) (2025-08-21)
### Bug Fixes:
* Respect resource pagination limits (#108) by kik4444
* eliminate chat_live compile warning (#107) by [@andyl](https://github.com/andyl)
* log the action name (#102) by [@barnabasJ](https://github.com/barnabasJ)
### Improvements:
* move permissions check of tools until after appropriate filtering (#104) by [@jgwmaxwell](https://github.com/jgwmaxwell)
* Add default adapter for ChatGoogleAI (#99) by [@mylanconnolly](https://github.com/mylanconnolly)
## [v0.2.10](https://github.com/ash-project/ash_ai/compare/v0.2.9...v0.2.10) (2025-08-07)
### Bug Fixes:
* log the action name (#102) by [@barnabasJ](https://github.com/barnabasJ)
### Improvements:
* move permissions check of tools until after appropriate filtering (#104) by [@jgwmaxwell](https://github.com/jgwmaxwell)
* Add default adapter for ChatGoogleAI (#99) by [@mylanconnolly](https://github.com/mylanconnolly)
## [v0.2.9](https://github.com/ash-project/ash_ai/compare/v0.2.8...v0.2.9) (2025-07-22)
### Improvements:
* mark all fields as required by [@zachdaniel](https://github.com/zachdaniel)
* handle number constraints by [@zachdaniel](https://github.com/zachdaniel)
* Add on_tool_start and on_tool_end callbacks (#96) by [@bradleygolden](https://github.com/bradleygolden)
## [v0.2.8](https://github.com/ash-project/ash_ai/compare/v0.2.7...v0.2.8) (2025-07-17)
### Improvements:
* add typed struct example to usage rules & docs by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.7](https://github.com/ash-project/ash_ai/compare/v0.2.6...v0.2.7) (2025-07-17)
### Bug Fixes:
* separate custom_context from llm initialization in ash_ai.gen.chat (#88) by [@germanbottosur](https://github.com/germanbottosur)
## [v0.2.6](https://github.com/ash-project/ash_ai/compare/v0.2.5...v0.2.6) (2025-07-05)
### Bug Fixes:
* handle missing user module more gracefully by [@zachdaniel](https://github.com/zachdaniel)
* properly install usage rules by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.5](https://github.com/ash-project/ash_ai/compare/v0.2.4...v0.2.5) (2025-07-03)
### Improvements:
* support sub rules in usage rules tools by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.4](https://github.com/ash-project/ash_ai/compare/v0.2.3...v0.2.4) (2025-07-02)
### Bug Fixes:
* allow for a custom json_processor (#80) by [@TwistingTwists](https://github.com/TwistingTwists)
* changed chat-live message history order before adding it to langchain (#78) by srmico
* crash with embedded resource (#77) by [@nallwhy](https://github.com/nallwhy)
### Improvements:
* add documentation for tool private attribute behavior (#81) by marot
* add documentation for tool private attribute behavior by marot
* install usage rules better by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.3](https://github.com/ash-project/ash_ai/compare/v0.2.2...v0.2.3) (2025-06-25)
### Bug Fixes:
* unsafe usage in mdex (#73) by [@TwistingTwists](https://github.com/TwistingTwists)
### Improvements:
* update usage rules w/ more prompt actions by [@zachdaniel](https://github.com/zachdaniel)
* multi-provider support prerequisite - eliminate open api spex reliance (#64) by KasparKipp
* Support various additional prompt formats (#72) by [@TwistingTwists](https://github.com/TwistingTwists)
## [v0.2.2](https://github.com/ash-project/ash_ai/compare/v0.2.1...v0.2.2) (2025-06-11)
### Bug Fixes:
* properly close connection after sending the endpoint by [@zachdaniel](https://github.com/zachdaniel)
### Improvements:
* use relative paths in usage rules MCP by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.1](https://github.com/ash-project/ash_ai/compare/v0.2.0...v0.2.1) (2025-06-11)
### Bug Fixes:
* fix installer waiting for input by [@zachdaniel](https://github.com/zachdaniel)
### Improvements:
* make usage rules display all and show file paths instead of by [@zachdaniel](https://github.com/zachdaniel)
## [v0.2.0](https://github.com/ash-project/ash_ai/compare/v0.1.11...v0.2.0) (2025-06-10)
### Features:
* Json Processor for providers that do not support json_schema or tool calling (#49) by [@TwistingTwists](https://github.com/TwistingTwists)
* improvement: Usage rules mcp integration (#60) by [Barnabas Jovanovics](https://https://github.com/barnabasJ)
### Bug Fixes:
* tasks: fix prompt typo (#62) by ChristianAlexander
* endpoint matching for url 'starting from' api.openai.com (#57) by [@TwistingTwists](https://github.com/TwistingTwists)
* fix oban option passing by [@zachdaniel](https://github.com/zachdaniel)
* require an explicit endpoint set by [@zachdaniel](https://github.com/zachdaniel)
* pass tenant to AshOban.run_trigger by [@zachdaniel](https://github.com/zachdaniel)
### Improvements:
* sync usage rules on project creation by [@zachdaniel](https://github.com/zachdaniel)
* more context in error messages (#56) by [@TwistingTwists](https://github.com/TwistingTwists)
* When using Adapter.CompletionTool (for anthropic) add the cache_control (#51) by Rodolfo Torres
* more realistic handling of example generation (#50) by [@TwistingTwists](https://github.com/TwistingTwists)
## [v0.1.11](https://github.com/ash-project/ash_ai/compare/v0.1.10...v0.1.11) (2025-06-04)
### Improvements:
* adapters for prompt-backed actions
* add completion tool adapter, infer it from anthropic
## [v0.1.10](https://github.com/ash-project/ash_ai/compare/v0.1.9...v0.1.10) (2025-05-30)
### Bug Fixes:
* use after_action instead of after_transaction to afford atomic_updates (#43)
## [v0.1.9](https://github.com/ash-project/ash_ai/compare/v0.1.8...v0.1.9) (2025-05-27)
### Bug Fixes:
* remove unnecessary source type from generated chat code
## [v0.1.8](https://github.com/ash-project/ash_ai/compare/v0.1.7...v0.1.8) (2025-05-27)
### Improvements:
* overhaul ash_ai.gen.chat to store tool calls
* make the dev mcp path configurable (#38)
## [v0.1.7](https://github.com/ash-project/ash_ai/compare/v0.1.6...v0.1.7) (2025-05-21)
### Improvements:
* Add usage rules for Ash AI
## [v0.1.6](https://github.com/ash-project/ash_ai/compare/v0.1.5...v0.1.6) (2025-05-21)
### Improvements:
* Rename package_ruels to usage_rules
## [v0.1.5](https://github.com/ash-project/ash_ai/compare/v0.1.4...v0.1.5) (2025-05-21)
### Bug Fixes:
* properly display generators, add new usage-rules.md dev tool
### Improvements:
* add `ash_ai.gen.package_rules` task to create a rules file
## [v0.1.4](https://github.com/ash-project/ash_ai/compare/v0.1.3...v0.1.4) (2025-05-20)
### Bug Fixes:
* Replace doc with description (#36)
## [v0.1.3](https://github.com/ash-project/ash_ai/compare/v0.1.2...v0.1.3) (2025-05-20)
### Bug Fixes:
* use `description` not `doc`
## [v0.1.2](https://github.com/ash-project/ash_ai/compare/v0.1.1...v0.1.2) (2025-05-20)
### Bug Fixes:
* improve chat ui heex template
* don't reply to the initialized notification (#35)
### Improvements:
* update chat heex template. (#33)
## [v0.1.1](https://github.com/ash-project/ash_ai/compare/v0.1.0...v0.1.1) (2025-05-14)
### Bug Fixes:
* more fixes for gen.chat message order
* properly generate chat message log
### Improvements:
* fix update pre_flight permission request for tools
## [v0.1.0](https://github.com/ash-project/ash_ai/compare/v0.1.0...v0.1.0) (2025-05-14)
### Bug Fixes:
* always configure chat queues
* Set additionalProperties to false in parameter_schema (#16)
* Fix load opt not working (#12)
* don't pass nil input in function/4 (#8)
* Fix schema type of actions of Options (#5)
* use `:asc` to put lowest distance records at the top
* use correct ops in vector before action
* use `message` instead of `reason`
### Improvements:
* add `mix ash_ai.gen.mcp`
* dev tools MCP
* remove vector search action
* Add an MCP server support
* support tool-level descriptions
* better name trigger
* use bulk actions for update/destroy
* first draft of `mix ash_ai.gen.chat` (#19)
* allow read actions to be aggregated in addition to run
* set up CI, various fixes and refactors
* Add aggregates to filter properties (#15)
* Add async opt to Tool
* Add load opt to tool (#9)
* Add tenant to opts of setup_ash_ai/2 (#4)
* add installer
* add tenants to action calls in functions
* add `:manual` strategy
* allow specifying tools by name of tool
* strict modes & other various improvements
* make embedding model parameterizable
* remove unnecessary deps, use langchain
* make embedding models for arbitrary vectorization
* use configured name for tools
* make the DSL more `tool` centric
* add vectorize section
