Providers
Provider setup guide for Coven Code, including Anthropic, Codex, Claude CLI import, model selection, rate limits, and credential boundaries.
2 min read
Coven Code currently supports Anthropic and Codex provider paths. You can connect from the startup provider prompt, the /connect picker inside the TUI, or provider-specific shell commands.
Provider Setup Prompt
On first launch, if no credentials are configured, Coven Code opens a provider setup prompt:
| Choice | Meaning |
|---|---|
1 | Start the best available Claude path, preferring local Claude CLI import when present and falling back to API key setup. |
2 | Start Codex browser login. |
Enter | Open the full /connect picker. |
Esc | Skip setup for now. |
You can always run /connect later.
Anthropic
Use Anthropic when you want Claude models through Coven Code's direct provider client. You can provide an API key:
export ANTHROPIC_API_KEY=<your-key>
coven-code --provider anthropicOr use the in-app /connect flow. For environments that support Coven Code's Anthropic OAuth client, coven-code auth login can store an OAuth account.
Claude CLI Import
Claude CLI import means Coven Code imports usable local Claude Code credentials into a Coven Code account profile. It is a credential import path, not a request proxy through the official claude CLI process.
Import is not the same as running Claude Code
When a session uses an imported Claude CLI account, Coven Code still sends requests through its own Anthropic provider client. Headers, retry behavior, queueing, and provider rate-limit buckets can differ from the official Claude Code CLI.
This distinction matters when debugging 429 rate limits: re-importing the same Claude CLI credentials will not create a new quota pool. Wait for provider limits to reset, choose a lighter model, switch providers, or run the official Claude Code CLI outside Coven Code when you specifically need that execution path.
Codex
Use Codex when you want the OpenAI Codex provider path through a ChatGPT/Codex login:
coven-code codex login
coven-code --provider codexInside the TUI, /connect can start the Codex login flow and /model can switch among available Codex models.
Model And Effort Selection
Use /model in the TUI to change models for the active provider. Use the left and right arrow controls in that picker, or the /effort command, to change reasoning effort where the model supports it.
From the shell:
coven-code --provider anthropic --model claude-sonnet-4-6
coven-code --provider codex --model gpt-5-codex
coven-code --effort highYou can list the model catalog with:
coven-code models
coven-code models anthropic --refresh
coven-code models codex --jsonProvider Boundary
Coven Code is local-first, but providers are still provider-owned services:
- API keys and OAuth records are local account state, not OpenCoven hosted accounts.
COVEN_CODE_PROVIDERselects the default provider.COVEN_CODE_API_BASEcan override the API base URL for compatible provider endpoints.- Model requests go to the selected provider; daemon requests go to Coven over the local socket.
For daemon-level harness auth, see Harness Provider Auth. For general auth assumptions, see Authentication Reference.
Last updated on