Skip to content

providers.md

Latest commit

 

History

History
185 lines (146 loc) · 4.85 KB

providers.md

File metadata and controls

185 lines (146 loc) · 4.85 KB

Providers & Models

BB-Agent supports multiple LLM providers out of the box.

Supported Providers

Provider Auth Method Models
Anthropic OAuth or ANTHROPIC_API_KEY Claude Opus, Sonnet, Haiku
OpenAI OAuth or OPENAI_API_KEY GPT-4o, GPT-4.1, o1, o3, o4-mini
GitHub Copilot OAuth/device flow or GH_COPILOT_TOKEN Copilot chat models
Google GOOGLE_API_KEY Gemini 2.5 Pro, Flash
Groq GROQ_API_KEY Llama, Mixtral
xAI XAI_API_KEY Grok
OpenRouter OPENROUTER_API_KEY 100+ models
Custom Configurable Any OpenAI-compatible API

Authentication

OAuth Login (Anthropic, OpenAI, GitHub Copilot)

bb login anthropic        # Opens browser for OAuth
bb login openai-codex     # Opens browser for OAuth
bb login github-copilot   # GitHub device flow + Copilot token exchange

For GitHub Copilot, bb now supports:

  • stored authority-aware configuration (github.com or GitHub Enterprise Server domain)
  • GitHub device/browser auth flow
  • GitHub OAuth token persistence in auth.json
  • Copilot runtime token exchange via GitHub's Copilot token endpoint
  • Copilot runtime token refresh by re-exchanging the saved GitHub OAuth session when GITHUB_COPILOT_CLIENT_SECRET (or GH_COPILOT_CLIENT_SECRET) is provided
  • /models validation and cached Copilot model discovery
  • Copilot auth/session visibility in /session

Current limitations:

  • Copilot request behavior is wired through the OpenAI-compatible runtime path and may still need endpoint/header adjustments for some models or enterprise installations
  • Enterprise endpoint behavior still needs more real-world validation

API Key Login

bb login google         # Prompts for API key
bb login groq
bb login xai
bb login openrouter

Environment Variables

Set directly without bb login:

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."
export GROQ_API_KEY="..."
export XAI_API_KEY="..."
export OPENROUTER_API_KEY="..."
export GH_COPILOT_TOKEN="..."                 # Direct Copilot runtime token
export GITHUB_COPILOT_TOKEN="..."             # Equivalent env fallback
export GITHUB_COPILOT_CLIENT_SECRET="..."     # Optional: only needed for GitHub OAuth refresh support

If you do not set GITHUB_COPILOT_CLIENT_SECRET, GitHub Copilot sign-in still works, but expired GitHub OAuth sessions must be refreshed by logging in again.

Check Status

bb login    # Shows ✓/✗ for each provider

Selecting a Model

CLI Flags

bb --model sonnet                                # Fuzzy match
bb --model claude-sonnet-4-20250514              # Exact model ID
bb --model anthropic/claude-sonnet-4-20250514    # Provider/model
bb --model sonnet:high                           # Model with thinking level
bb --provider google --model gemini-2.5-flash    # Explicit provider

Thinking Levels

For models that support extended thinking:

bb --model sonnet:high      # High thinking budget
bb --model sonnet:medium    # Medium (default)
bb --model sonnet:low       # Low
bb --model sonnet:off       # No extended thinking
bb --thinking high          # Set thinking separately

List Available Models

bb --list-models            # List all models
bb --list-models sonnet     # Search/filter
bb --list-models groq       # Models from a provider

In-Session Model Switching

Press Ctrl+P to cycle through models, or use:

/model sonnet
/model gpt-4o
/model openai/gpt-4o
/model openai:gpt-4o
/model sonnet:high
/model anthropic/claude-sonnet-4-20250514:low

/model now accepts common provider/model and thinking-suffix formats directly during a conversation.

Default Model

In settings.json:

{
  "default_provider": "anthropic",
  "default_model": "claude-sonnet-4-20250514",
  "default_thinking": "medium"
}

Custom Models

Add models that aren't in the built-in registry:

{
  "models": [
    {
      "id": "llama3-70b",
      "name": "Llama 3 70B (local)",
      "provider": "ollama",
      "api": "openai",
      "base_url": "http://localhost:11434/v1",
      "context_window": 8192,
      "max_tokens": 4096,
      "reasoning": false
    }
  ]
}

Custom Providers

Define entirely new providers:

{
  "providers": [
    {
      "name": "my-corp",
      "base_url": "https://llm.internal.corp.com/v1",
      "api_key_env": "CORP_LLM_KEY",
      "api": "openai",
      "headers": {
        "X-Team": "engineering"
      }
    }
  ]
}

Then use:

bb --provider my-corp --model our-model

API Types

The api field determines the request/response format:

Value Compatible With
openai OpenAI, Groq, xAI, OpenRouter, Ollama, vLLM, LiteLLM
anthropic Anthropic
google Google Gemini