Skip to content

providers.md

Latest commit

 

History

History
82 lines (59 loc) · 2.46 KB

providers.md

File metadata and controls

82 lines (59 loc) · 2.46 KB
Raw

Providers

OpenLLMAuth resolves provider/model references through src/open_llm_auth/auth/manager.py and the built-in catalog in src/open_llm_auth/provider_catalog.py.

Provider Reference Format

Prefer explicit provider/model references:

provider/model

Examples:

openai/gpt-5.2
anthropic/claude-sonnet-4-6
kimi-coding/k2p5
ollama/llama3.1

Bare model IDs are accepted only when provider inference is unambiguous or a small built-in heuristic can safely infer the provider.

Built-In Provider Families

The catalog includes these provider families:

  • OpenAI-compatible: openai, google, moonshot, openrouter, together, huggingface, mistral, venice, nvidia, qianfan, volcengine, byteplus, kilocode, opencode, vercel-ai-gateway, litellm, ollama, and vllm.
  • Anthropic-compatible: anthropic, kimi-coding, minimax, minimax-portal, xiaomi, synthetic, and cloudflare-ai-gateway.
  • Codex and coding surfaces: openai-codex, claude-cli, and codex-cli.
  • GitHub Copilot: github-copilot.
  • AWS Bedrock Converse: amazon-bedrock.
  • Local agent runtime bridge: agent_bridge and the agent alias.

Credential Resolution

For a requested model, the provider manager resolves credentials in this order:

  1. Explicit preferred profile, if supplied.
  2. Configured auth-order list for the provider.
  3. Discovered profiles for the provider.
  4. Provider-specific environment variables.
  5. Provider config api_key.
  6. Special provider paths such as AWS SDK or CLI-backed providers.

OAuth and Device Login Helpers

OpenLLMAuth includes CLI helpers for providers that need interactive auth:

uv run open-llm-auth auth login-openai-codex
uv run open-llm-auth auth login-github-copilot

You can also add token and OAuth profiles manually:

uv run open-llm-auth auth add-token PROVIDER --profile default
uv run open-llm-auth auth add-oauth PROVIDER --profile default

Local Providers

ollama, vllm, and litellm point at local OpenAI-compatible endpoints by default. They do not become active simply because they are listed in the catalog; the backing local service must be running and the provider must resolve under the current auth/config rules.

Agent Bridge Provider

agent_bridge is a local runtime bridge rather than a remote model provider. It can proxy normal chat calls and lifecycle task calls to a compatible local agent runtime.

See Agent Bridge Integration.