Contributing Providers

Contributing a Provider

Adding a new AI provider to AI-Protocol makes it instantly available across all runtimes (Rust, Python, and any future implementations).

Steps

1. Research the Provider API

Document the following about the provider:

• Base URL and chat endpoint path
• Authentication method (Bearer token, API key header, etc.)
• Request parameter format
• Streaming response format (SSE, NDJSON, custom)
• Error response structure
• Available models and their capabilities

2. Create the Provider Manifest

Create v1/providers/<provider-id>.yaml:

id: <provider-id>
name: "<Provider Name>"
protocol_version: "1.5"

endpoint:
  base_url: "https://api.example.com/v1"
  chat_path: "/chat/completions"

auth:
  type: bearer
  token_env: "<PROVIDER_ID>_API_KEY"

parameter_mappings:
  temperature: "temperature"
  max_tokens: "max_tokens"
  stream: "stream"
  tools: "tools"

streaming:
  decoder:
    format: "sse"
    done_signal: "[DONE]"
  event_map:
    - match: "$.choices[0].delta.content"
      emit: "PartialContentDelta"
      extract:
        content: "$.choices[0].delta.content"

error_classification:
  by_http_status:
    "401": "authentication"
    "429": "rate_limited"
    "500": "server_error"

capabilities:
  streaming: true
  tools: true
  vision: false

3. Add Models

Create or update v1/models/<family>.yaml:

models:
  example-model:
    provider: <provider-id>
    model_id: "example-model-v1"
    context_window: 128000
    capabilities: [chat, streaming, tools]
    pricing:
      input_per_token: 0.000001
      output_per_token: 0.000002

4. Validate

npm run validate

This checks your manifest against the JSON Schema and reports any errors.

5. Build

npm run build

This compiles your YAML to JSON in the dist/ directory.

6. Submit a Pull Request

• Fork the repository
• Create a branch
• Add your provider manifest and model entries
• Ensure validation passes
• Submit a PR with documentation about the provider

Validation Rules

The JSON Schema enforces:

• Required fields (id, endpoint, auth, parameter_mappings)
• Valid formats for URLs, environment variable names
• Correct structure for streaming configuration
• Valid error classification types
• Capability flags as booleans

Tips

• Use the OpenAI-compatible format if the provider follows the OpenAI API structure — many providers do (Groq, Together AI, DeepSeek)
• Test streaming configuration carefully — this is where most provider differences exist
• Include capabilities flags accurately — runtimes use them for pre-flight validation