- Vercel AI SDK v5 support - This version requires
@ai-sdk/providerv2.0.0 and@ai-sdk/provider-utilsv3.0.1 - Previous versions of Vercel AI SDK (v4) are no longer supported
- Updated
aidev dependency to v5.0.0
- Refactored
langtail-language-model.tsfor AI SDK v5 compatibility - Updated
convert-to-openai-chat-messages.tsfor new message format - Updated
map-langtail-finish-reason.tsfor v5 finish reasons - Updated
openai-prepare-tools.tsfor v5 tool format - Removed deprecated
map-openai-chat-logprobs.ts - Added comprehensive test coverage for language model and message conversion
- Add GoogleGeminiV1 to ReasoningFormat and update related tests
- Improved shape of reasoning_details values - not using native enum
- Add support for reasoning_details (openrouter specific) returned in providerMetadata in the vercel ai sdk provider
- Add support for tool results as message type arrays with images in Vercel AI SDK integration
- Tool results can now return rich content arrays containing text and images instead of just strings
- Maintains backward compatibility with existing string-based tool results
- Updated
ChatCompletionToolMessageParamto support both string and array content types
- Add support for message.reasoning in schema
- Add simpleHash function to generate unique IDs for tool calls in LangtailChatLanguageModel to prevent collisions, particularly with Google Gemini 2.5
- Update tool call ID logic to incorporate hashed arguments
- support null in reasoning (returned by openrouter sometimes)
- Fixed streaming support for Claude 3.7 Sonnet reasoning format
- Enhanced Zod schema for reasoning field to handle a wider variety of response formats
- Improved type handling for reasoning in Vercel AI SDK integration
- Added cache_enabled support to message schema
- Added Anthropic cache control support to Vercel AI SDK integration
- Updated dependencies: '@ai-sdk/provider' to 1.1.1 and '@ai-sdk/provider-utils' to 2.2.5
- Added 'ai' package as dev dependency
- Added playground examples for Vercel integration with cache control
- Fixed reasoning for Claude thinking
- Added changes from 0.13.11 and 0.13.10
- Reasoning data support
- Fix
finishReasonfor Anthropic and Google providers in Vercel AI SDK integration
- Update
openaiversion - Fix tool calls issues for Vercel AI SDK integration
- Remove
useChatStream(moved to separate repo)
- Add
max_thinking_tokenssupport (Anthropic enhanced thinking)
- Add reasoning_effort support
- Fix for browser runtime
- New Handlebars helpers: last, toJSON, formatMessage (more variants)
- Fix react peerDependency version syntax #73
- Keep the tool_calls in final message callback #72
- Option to append template to the end in
getOpenAIBody. - Fix IntelliSense for tools when using Vercel AI SDK.
- add parallelToolCalls option to pass it to the API
- Support for Langtail hosted tools
- Fix: don't send
stopparameter if it's empty (this causes validation error in OpenAI in some cases) - Add schemas to entrypoints
- Fix CommonJS support
- Add
promptparameter to OpenAI proxy
- Fix generating quoted variable values with
- BREAKING CHANGE: rename vercelAi -> vercel-ai
- BREAKING CHANGE: rename
LangtailNodeclass toLangtail - add
langtail/openaiwithOpenAIProxyclass - BREAKING CHANGE: remove
Langtail.chat, useOpenAIProxyinstead
- BREAKING CHANGE: remove
langtail.buildmethod, usegetOpenAIBodyinstead
- BREAKING CHANGE: remove
dist/from import paths
- IntelliSense for variables
- useChatStream react hook
- export stream helpers
chatStreamToRunnerandassistantStreamToRunner
- Langtail types generator
- list deployments EP
- Langtail tools generator for Vercel AI SDK
- remove zod openapi dependency
- internal changes #24
- provider for Vercel AI (https://github.com/vercel/ai)
- fix next.js compatibility
- add ability to
getandbuildto run langtail prompts without using a proxy
- invoke
environmentparameter now defaults to production - start sending user-agent header #10
- Small README update - move short description above CI badges
- Rename to just
langtailNPM package
- Fixed an issue when passing
doNotRecord: falseas option, support for metadata fields
- Allow Node 18 (engines.node set to >= 18)
- First publicly used version 🎉🎉🎉