Providers
Choose a credential method, connect a provider, and verify the models your gateway can use.
A provider is a registry upstream such as OpenAI, Claude Code, or Ollama Local. A connection is one saved credential for that provider; you can keep several and DurinDoor will try another account when the current one fails. A provider node is a custom OpenAI-compatible or Anthropic-compatible endpoint you add yourself, with its own prefix and base URL.
Open Dashboard, then Providers. Registry cards group OAuth, API key, cookie, free, and custom nodes. Account fallback stays on the same provider and model; combo fallback is a different chain.
| Term | Meaning |
|---|---|
| Provider | A registry id (openai, claude, ollama-local) or a node id (openai-compatible-chat-...). |
| Connection | One credential row: OAuth tokens, API key, or cookie. Priority decides fill-first order. |
| Provider node | A user-added OpenAI-compatible, Anthropic-compatible, or embedding endpoint. |
| Account fallback | Next connection for the same provider and model after a fallback-worthy error. |
| Combo fallback | Next model in a combo after every account for the current member has failed. |
| Credential | Typical providers | Where it is documented |
|---|---|---|
| OAuth | Claude Code, Codex, Gemini CLI, GitHub | Connecting accounts |
| Cookie | Grok Web, Kimi Web, and other webCookie ids | Web cookie providers |
| API key | OpenAI, Anthropic, Groq, Azure | API key providers |
| Compatible node | Your own /v1 gateway | OpenAI-compatible nodes |
| Local / no-auth | Ollama, LM Studio, Horde | Free and local |
Add and test a connection
- Open Providers and choose a visible provider card. For a custom gateway, use Add OpenAI Compatible or Add Anthropic Compatible. Embedding-only nodes are under Media Providers.
- Follow the credential guide for OAuth, cookies, an API key, or a local server.
- Save the connection and check its active state and model selection.
- List models from
GET /v1/models, then send a first request. - Confirm the selected provider and account on Usage.
A saved credential or successful probe does not prove chat, streaming, tools, or media work. The catalog includes registry metadata, including hidden or unsupported entries. The dashboard and a real request establish what your instance can use.
If an account is unavailable, routing can try another eligible connection for the same provider and model. A combo can then try another model. Priority controls fill-first order; other strategies are optional. Scope client keys when only selected accounts should be reachable.
Connecting accounts
OAuth, cookies, token refresh, and account priority.
API key providers
Paste a key, env fallbacks, and key-to-account scoping.
OpenAI-compatible nodes
Custom OpenAI, Anthropic, and embedding endpoints.
Free and local
No-auth, Ollama, LM Studio, cookies, and local devices.
Local router providers
Self-hosted OpenAI-compatible IDs and metadata-only entries.
Web cookie providers
Copy a browser session cookie and what each site expects.
Provider catalog
Generated list of registry ids, auth types, and model counts.
Provider guides
Some providers need more than the generic flow. These pages cover sign-in, imports, quota, and the errors specific to each one.
Claude Code
Claude subscription OAuth, client compatibility, and five-hour and weekly quota.
OpenAI Codex
ChatGPT OAuth on port 1455, bulk import, fingerprint mode, and reauth quarantine.
Antigravity
Google OAuth client variables, project onboarding, the agy id, and MITM mode.
Kiro
Builder ID, Identity Center, social login, API key, and IDE token import.
Cursor
Import the Cursor IDE session, auto-detect paths, and no refresh.
Kimi Web
Copy localStorage tokens, choose models, and refresh an expired session.
Hidden and retired providers
Providers you can no longer add and ids with no executor yet.