feat: Add Mistral AI provider support

[Esashiero]

Summary

This PR adds Mistral AI as a supported LLM provider in PicoClaw.

Changes

1. Uses /v1/conversations API endpoint - Different rate limits than chat completions, accepts "latest" model names

2. Request format conversion:

   • messages → inputs
   • System prompts → instructions field
   • max_tokens, temperature → completion_args

3. Tool support:

   • Uses Mistral's built-in web_search tool
   • Converts tool results to user messages for feedback loop

4. Response parsing:

   • Handles array content format when tools are used
   • Parses tool execution outputs

Usage

{
  "agents": {
    "defaults": {
      "provider": "mistral",
      "model": "mistral-large-latest"
    }
  },
  "providers": {
    "mistral": {
      "api_key": "your-mistral-api-key"
    }
  }
}

Testing

• Simple queries work (no tools)
• Web search queries work (with tools) - full flow verified

[Leeaandrob]

@Zepan Mistral AI is a good provider to support. However, PR #213 (Provider Refactor — roadmap #283, In Progress) will restructure how providers are organized.

Recommendation: Wait for #213 to merge first. The refactor groups providers by protocol (OpenAI-compatible, Anthropic, etc.) — Mistral would likely fit into the OpenAI-compatible family, making this PR much simpler after the refactor.

[yinwm]

Thanks for the contribution! This PR adds valuable Mistral-specific features (native /v1/conversations endpoint, built-in web_search).

However, this PR needs to be reworked due to recent architecture changes.

Architecture Update:

PR #492 (Provider Refactoring) has been merged on 2026-02-20, introducing a protocol-based model_list configuration. See the design doc: provider-refactoring.md

Current Status:

Mistral is not yet registered as a protocol. However, users can already use Mistral via OpenAI-compatible API:

{
  "model_list": [
    {
      "model_name": "mistral",
      "model": "openai/mistral-large-latest",
      "api_base": "https://api.mistral.ai/v1",
      "api_key": "your-key"
    }
  ]
}

Suggested Implementation for Native Mistral API:

To add native /v1/conversations support with web_search, please rebase and restructure as follows:

1. Create pkg/providers/mistral/provider.go:

   • Implement MistralProvider struct with Chat() method
   • Handle /v1/conversations endpoint
   • Convert message format: messages → inputs, system prompt → instructions
   • Support built-in web_search tool

2. Register protocol in pkg/providers/factory_provider.go:

   case "mistral":
       apiBase := cfg.APIBase
       if apiBase == "" {
           apiBase = "https://api.mistral.ai/v1"
       }
       return NewMistralProvider(cfg.APIKey, apiBase), modelID, nil

   Also add to getDefaultAPIBase():

   case "mistral":
       return "https://api.mistral.ai/v1"

3. Example configuration:

   {
     "model_list": [
       {
         "model_name": "my-mistral",
         "model": "mistral/mistral-large-latest",
         "api_key": "your-key"
       }
     ]
   }

You can refer to pkg/providers/anthropic/provider.go as a reference for the structure.

Would you like to rebase and update this PR with the new approach? Happy to help review once updated!

[nikolasdehor]

PR #541 implements native Mistral AI provider support, which addresses this feature request. Linking for visibility.