new-provider

New Model Provider

Create a new model provider following established patterns.

When to Use

• Adding support for a new LLM provider (Anthropic, Mistral, etc.)
• Creating a custom model provider
• Understanding model provider structure

Available Operations

1. Scaffold: Create directory structure from template
2. Configure: Set up tool.gpt and credentials
3. Implement: Write provider-specific code
4. Test: Validate the implementation

Instructions

Step 1: Gather Information

Required:

• Provider name (e.g., "anthropic", "mistral")
• API base URL
• Authentication method (API key, OAuth)

Step 2: Copy Template

cp -r obot-tools/openai-model-provider obot-tools/{name}-model-provider

Step 3: Update tool.gpt

See references/tool-gpt-template.md for format:

• Change Name: to {Name} Model Provider
• Update Description:
• Update credential if different auth method

Step 4: Update Go Code

• Rename package
• Update API endpoint
• Adjust request/response mapping if API differs

Step 5: Update index.yaml

Add to modelProviders: section:

- name: {name}-model-provider
  description: {Name} model provider for Obot
  path: ./{name}-model-provider

Step 6: Test

cd obot-tools/{name}-model-provider
export OBOT_{NAME}_MODEL_PROVIDER_API_KEY=...
go run . validate

Resources

Load these Level 3 resources as needed:

• references/tool-gpt-template.md - GPTScript template for model providers
• references/provider-patterns.md - Common provider implementation patterns
• assets/main.go.tmpl - Go main.go template

Examples

Example: Anthropic Provider

User asks: "Create an Anthropic model provider"

Response:

1. Copy openai-model-provider
2. Update Name to "Anthropic Model Provider"
3. Update API endpoint to api.anthropic.com
4. Adjust message format for Claude API
5. Test with validate command

Example: Mistral Provider

User asks: "Add Mistral AI support"

Response:

1. Copy openai-model-provider (Mistral is OpenAI-compatible)
2. Update Name to "Mistral Model Provider"
3. Update base URL to api.mistral.ai
4. Minimal code changes (OpenAI compatible)
5. Test with validate command

Checklist

Before committing:

• tool.gpt has correct Name, Description, Tools
• Environment variable follows pattern: OBOT_{NAME}_MODEL_PROVIDER_*
• Implements OpenAI-compatible /v1/chat/completions endpoint
• Error responses are user-friendly
• Added to index.yaml
• Tested with go run . validate

Notes

• Most providers follow OpenAI-compatible format
• Check if provider has OpenAI-compatible endpoint first
• Environment variables follow pattern: OBOT_{NAME}_MODEL_PROVIDER_*