Providers And Compatibility Reference
To add, edit, or test a model service, go directly to AI Providers. This page is a reference for how provider families and client interfaces relate.
Common Providers
| Provider type | Common use | Related pages |
|---|---|---|
| Codex | Codex CLI, accounts, and quota | AI Providers, Accounts, Inspection |
| Claude | Claude Code and compatible calls | AI Providers, Accounts, Monitoring |
| OpenAI-compatible | Relays, self-hosted, or compatible services | AI Providers, Model Prices, Usage Analytics |
| Gemini / Vertex | Google models and project credentials | AI Providers, OAuth, Accounts |
| xAI / Grok | API key or OAuth accounts | AI Providers, Accounts, Inspection |
| Muse / Meta | Device Flow account or meta-api-key inference provider | AI Providers, OAuth, Accounts |
When adding a provider, confirm four things first: base URL, authentication method, model names used by clients, and the account or auth file binding.
Muse / Meta requires an extra credential distinction: OAuth Device Flow produces a DCA for account authentication and quota reads, while model inference uses a separate meta-api-key. Full support requires CPA v7.3.4+.
When Requests Fail
- Run an available model or key test from AI Providers.
- Check whether the auth file is disabled or out of quota.
- Send a low-cost real request.
- Read the status code and sanitized failure summary in Monitoring.
- Use Logs only after those checks.
Advanced: compatibility APIs and reverse proxy
Common client interfaces include:
/v1/...for OpenAI-compatible clients./v1beta/...for Gemini-compatible clients./backend-api/codex/...for Codex CLI.- Provider callback paths for OAuth login.
Model requests must go to CPA, not CPAMP. For same-domain routing, see Reverse Proxy.
Model prices affect CPAMP local cost estimates only. They do not change CPA routing or provider billing.