Custom tool call parser for AI SDK

Note

Depends on AI SDK v5 release, if you wish to use it on v4, please pin the package version to 1.0.0

Allows tool calls to be used in the AI ​​SDK framework regardless of the model.

Why This Exists

Many self‑hosted or third‑party model endpoints (vLLM, MLC‑LLM, Ollama, OpenRouter, etc.) don’t yet expose the OpenAI‑style tools parameter, forcing you to hack together tool parsing.
This project provides a flexible middleware that:

• Parses tool calls from streaming or batch responses
• Supports Hermes and Gemma formats
• Llama, Mistral, and JSON formats are coming soon
• Gain complete control over the tool call system prompt.

Installation

pnpm install @ai-sdk-tool/parser

---

Example: Gemma3‑Style Middleware

See examples/core/src/00-stream-tool-call.ts for the full demo:

// filepath: examples/core/src/00-stream-tool-call.ts
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';
import { wrapLanguageModel, streamText } from 'ai';
import { gemmaToolMiddleware } from '@ai-sdk-tool/parser';

const openrouter = createOpenAICompatible({ /* ... */ });

async function main() {
  const result = streamText({
    model: wrapLanguageModel({
      model: openrouter('google/gemma-3-27b-it'),
      middleware: gemmaToolMiddleware,
    }),
    system: 'You are a helpful assistant.',
    prompt: 'What is the weather in my city?',
    maxSteps: 4,
    tools: {
      get_location: { /* ... */ },
      get_weather: { /* ... */ },
    },
  });

  for await (const part of result.fullStream) {
    // ...handling text-delta and tool-result...
  }
}

main().catch(console.error);

---

🤝 Contributing

• Feel free to open issues or PRs—especially for new model formats.
• See CONTRIBUTING.md for guidelines.