@friendliai/ai-provider
Version:
<!-- header start --> <p align="center"> <img src="https://huggingface.co/datasets/FriendliAI/documentation-images/resolve/main/model-card-assets/friendliai.png" width="100%" alt="FriendliAI Logo"> </p> <!-- header end -->
396 lines (224 loc) • 8.93 kB
Markdown
@friendliai/ai-provider
# 1.1.15
## Patch Changes
- b1d75a9: Deprecate DeepSeek-V3.1
# 1.1.14
## Patch Changes
- 68f6baf: Add gemma-4-31B-it to FSE model list
# 1.1.13
## Patch Changes
- f558309: Deprecate Qwen3-30B-A3B
# 1.1.12
## Patch Changes
- ec80707: Deprecate GLM-4.7
# 1.1.11
## Patch Changes
- 8b25516: Deprecate MiniMax-M2.1
# 1.1.10
## Patch Changes
- e177dc2: Add GLM-5.1 to FSE model list
# 1.1.9
## Patch Changes
- 6e806c2: update whisper-large-v3
- 6e806c2: update get-available-model
# 1.1.8
## Patch Changes
- 342afa5: add k-exaone-236b-a23b model
# 1.1.7
## Patch Changes
- bbdecbb: deprecate fse models
# 1.1.6
## Patch Changes
- d56dcfc: add deepseek v3.2 in FSE
# 1.1.5
## Patch Changes
- f09ec79: Remove EXAONE-4.0.1 from FSE model list
# 1.1.4
## Patch Changes
- f97fb70: Remove K-EXAONE from FSE model list
- 3664158: Improve AI SDK compatibility in @friendliai/ai-provider by updating core SDK dependencies, wrapping Friendli JSON envelope chat errors as APICallError, and aligning chat usage and tool streaming events.
# 1.1.3
## Patch Changes
- a4954be: Add MiniMax-M2.5 to FSE model list
# 1.1.2
## Patch Changes
- f2f1abf: Add GLM-5 to FSE model list
# 1.1.1
## Patch Changes
- 5636c29: update-fse-models
# 1.1.0
## Minor Changes
- 7dbbdb5: Add interleaved thinking support for reasoning models like MiniMax-M2.1, GLM-4.7. Assistant messages with reasoning parts are now converted to include `reasoning_content` field in API requests, enabling multi-turn conversations to maintain reasoning context.
# 1.0.4
## Patch Changes
- e7ace16: Add `MiniMaxAI/MiniMax-M2.1` and remove `MiniMaxAI/MiniMax-M2` from FSE supported model list
- c09d857: remove deepseek r1 0528 from fse models
# 1.0.3
## Patch Changes
- df0f0b8: Fix finishReason format to match AI SDK v6 requirements and add proper text/reasoning event lifecycle
# 1.0.2
## Patch Changes
- fca96fa: Fix AI SDK v6 streaming protocol compliance
This change fixes the streaming implementation to properly comply with the AI SDK v6 LanguageModelV3 protocol:
**Fixed Issues:**
- Added missing `text-start` and `reasoning-start` events at the beginning of content blocks
- Changed incorrect event types from `text-part-finish` to `text-end` and `reasoning-part-finish` to `reasoning-end`
- Implemented consistent ID tracking across the entire content block lifecycle
**Event Sequence (Before):**
```
text-delta → text-delta → text-part-finish (❌ incorrect)
```
**Event Sequence (After):**
```
text-start → text-delta → text-delta → text-end (✅ correct)
```
This resolves the "text part ... not found" error that occurs when using the provider with AI SDK v6.
# 1.0.1
## Patch Changes
- 49b9dc7: add new fse model
# 1.0.0
## Major Changes
- 4dc62a9: Entering the AI SDK v6 compatible migration
## Minor Changes
- 4dc62a9: Add Linkup integrated tool helper (`friendli.tools.linkupSearch`) for Tool Assisted API.
- 4dc62a9: ### Provider Tools Improvements
- **Renamed tool functions** for cleaner API:
- `webSearchBetaTool()` → `webSearch()`
- `webUrlBetaTool()` → `webUrl()`
- `mathCalendarBetaTool()` → `mathCalendar()`
- `mathStatisticsBetaTool()` → `mathStatistics()`
- `mathCalculatorBetaTool()` → `mathCalculator()`
- `codePythonInterpreterBetaTool()` → `codePythonInterpreter()`
- **Added streaming support for provider-executed tools**: Tool calls and results from Friendli's built-in tools (web search, calculator, etc.) are now properly streamed, allowing users to see tool execution progress in real-time.
- **Fixed tool type compatibility**: Provider tools now correctly implement the AI SDK `Tool` type with `inputSchema`, resolving type errors when using `friendli.tools.*` with `generateText` or `streamText`.
## Patch Changes
- 4dc62a9: Fix LanguageModelV3FinishReason type compatibility with AI SDK v6
Update finishReason handling to use the new object format `{ unified, raw }` instead of plain string, aligning with @ai-sdk/provider v3.0.0 type changes.
- 4dc62a9: Support Friendli serverless chat options (`topK` -> `top_k`, sampling providerOptions) and improve error propagation for HTTP and streaming tool-assisted responses.
# 1.0.0-beta.2
## Minor Changes
- 049179d: Add Linkup integrated tool helper (`friendli.tools.linkupSearch`) for Tool Assisted API.
## Patch Changes
- 049179d: Support Friendli serverless chat options (`topK` -> `top_k`, sampling providerOptions) and improve error propagation for HTTP and streaming tool-assisted responses.
# 1.0.0-beta.1
## Minor Changes
- 2ffbb78: ### Provider Tools Improvements
- **Renamed tool functions** for cleaner API:
- `webSearchBetaTool()` → `webSearch()`
- `webUrlBetaTool()` → `webUrl()`
- `mathCalendarBetaTool()` → `mathCalendar()`
- `mathStatisticsBetaTool()` → `mathStatistics()`
- `mathCalculatorBetaTool()` → `mathCalculator()`
- `codePythonInterpreterBetaTool()` → `codePythonInterpreter()`
- **Added streaming support for provider-executed tools**: Tool calls and results from Friendli's built-in tools (web search, calculator, etc.) are now properly streamed, allowing users to see tool execution progress in real-time.
- **Fixed tool type compatibility**: Provider tools now correctly implement the AI SDK `Tool` type with `inputSchema`, resolving type errors when using `friendli.tools.*` with `generateText` or `streamText`.
# 1.0.0-beta.0
## Major Changes
- 3044ef8: Entering the AI SDK v6 compatible migration
# 0.3.4
## Patch Changes
- 795d5bf: add minimax-m2 in FSE
# 0.3.3
## Patch Changes
- a6180ef: add deepseek v3.1 in FSE
# 0.3.2
## Patch Changes
- 0e44c15: Update FSE model list
# 0.3.1
## Patch Changes
- d6a924b: For convenience, parse_reasoning is set to true by default.
# 0.3.0
## Patch Changes
- 10ab7aa: add chat_template_kwargs option
- 10ab7aa: Added supported type model to friendli-provider
- 10ab7aa: bump v5 package latest
- 10ab7aa: dump deps (alpha.6)
- 10ab7aa: Initial build for ai sdk v5 support (broken Friendli compatibility)
- 10ab7aa: As of 25.08.19, the latest serverless list and model list are synchronized.
- 10ab7aa: enable regex on v5 (beta)
- 10ab7aa: Add getAvailableModels() for dynamic model discovery
ref: https://vercel.com/docs/ai-gateway/models-and-providers#dynamic-model-discovery
- 10ab7aa: enable support Tool Assisted API
# 0.3.0-beta.7
## Patch Changes
- ac374ab: Add getAvailableModels() for dynamic model discovery
ref: https://vercel.com/docs/ai-gateway/models-and-providers#dynamic-model-discovery
# 0.3.0-beta.6
## Patch Changes
- 0ed9e14: As of 25.08.19, the latest serverless list and model list are synchronized.
# 0.3.0-beta.5
## Patch Changes
- 40e93c0: bump v5 package latest
# 0.3.0-alpha.4
## Patch Changes
- be3a6bf: add chat_template_kwargs option
- 74cc1e2: dump deps (alpha.6)
# 0.2.7-alpha.2
## Patch Changes
- 07b0162: enable regex on v5 (beta)
- 3f15a05: enable support Tool Assisted API
# 0.2.7-alpha.1
## Patch Changes
- dda6520: Added supported type model to friendli-provider
# 0.2.7-alpha.0
## Patch Changes
- 5ec8524: Initial build for ai sdk v5 support (broken Friendli compatibility)
# 0.2.6
## Patch Changes
- 3e347dc: bump dependencies version
# 0.2.5
## Patch Changes
- 0db11c5: llama 3.1 70b removal
# 0.2.4
## Patch Changes
- 3be71b2: add deepseek r1 model
# 0.2.3
## Patch Changes
- 3085e03: update serverless model list
# 0.2.2
## Patch Changes
- b3e1fb1: Add Model Llama 3.3 70B Serverless
# 0.2.1
## Patch Changes
- edc5931: fix parallel tool call
# 0.2.0
## Minor Changes
- f262011: convert-openai-compatible-components
- dedicated, serverless automatic switching
- /v1/completion available
- Improved error parsing logic
- Update dependency packages
- Liquidate some of the legacy
- Other minor bug fixes
## Patch Changes
- bcdf26e: Convert regex input to RegExp type
# 0.1.25
## Patch Changes
- 894764d: fix model list
# 0.1.24
## Patch Changes
- 6b241b6: add exaone
# 0.1.23
## Patch Changes
- d51f04e: update dependencies, @ai-sdk/provider-utils@2.0.3
# 0.1.22
## Patch Changes
- 0fc7b03: Add Llama vision 11b support
# 0.1.21
## Patch Changes
- 89b09de: Built-in tool non-streaming support
# 0.1.20
## Patch Changes
- 7caa63e: ## Updated code to match ai sdk 4.0.10
- createFriendliAI -> createFriendli
- Dependency Update
- Updated tool call parsing logic
- Improved object creation logic
- Added regex output support
- delete "content_filter", "function_call" finish reason
# 0.1.19
## Patch Changes
- 868452c: inference.friendli.ai -> api.friendli.ai
# 0.1.18
## Patch Changes
- ba2f03e: First setup with turbo pack