Skip to content
smouter
Sign inStart free
Connect · CLI agents

Connect Codex CLI to Smouter

Point OpenAI’s Codex CLI at Smouter with a custom model provider in ~/.codex/config.toml.

What you need

  • Base URL: https://api.smouter.ai/v1 — the /v1 suffix matters; leaving it off is the most common setup failure.
  • API key: an sk-smouter-… secret from API keys. It is shown once at creation — if you lost it, create a new key.
  • Model id: copied exactly from the live catalog (or GET /v1/models). Ids are matched verbatim — no vendor prefixes, no renaming.

Point Codex CLI at Smouter

  1. Add a smouter provider to ~/.codex/config.toml and select it with model_provider.
  2. env_key names the environment variable Codex reads the key from — export it before launching.
  3. Keep wire_api = "chat". Smouter also serves the Responses API, but statelessly: Codex's responses mode can send previous_response_id, which a stateless surface rejects with 400. The chat wire avoids that entirely.
model = "<model-id>"model_provider = "smouter" [model_providers.smouter]name = "Smouter"base_url = "https://api.smouter.ai/v1"env_key = "SMOUTER_API_KEY"wire_api = "chat"

The Codex IDE extension reads the same ~/.codex/config.toml, so this provider works there too. The Codex built into the ChatGPT desktop app is tied to ChatGPT accounts and cannot use a custom provider.

Verify outside the tool first

Two commands separate a Smouter problem from a tool problem. If both succeed, your key, credit, and model id are fine — whatever remains is the tool's configuration.

# Which model ids can I use right now? (no auth needed)curl -s https://api.smouter.ai/v1/models | jq -r '.data[].id'

If it doesn't work

  • Base URL wrong or missing /v1. OpenAI-style tools need exactly https://api.smouter.ai/v1 — no trailing /chat/completions, no bare api.smouter.ai. Anthropic-style tools (Claude Code) use https://api.smouter.ai with no suffix.
  • 401 auth — the key was pasted wrong, expired, or revoked. Secrets are shown once; create a fresh key in API keys and paste it whole, including the sk-smouter- prefix.
  • 402 insufficient_quota — your wallet or a per-key cap can't cover the request. Top up in Billing or raise the key's cap.
  • 404 model_not_found — the id doesn't match the catalog. Copy it verbatim from the catalog; ids are case-sensitive and carry no openai/-style prefix (where a tool requires one, the tool strips it before sending).
  • 429 — a rate or tokens-per-minute window is exhausted. Honor retry-after; per-key TPM is adjustable in API keys.
  • Scoped key. If the key has a model allowlist, every model the tool may call — including fallbacks — must be on it. When in doubt, test with an unscoped key.
  • The tool's UI moved. Settings get renamed; the three constants don't. Any field asking for an OpenAI(-compatible) base URL, key, and model name takes the values above.

Still stuck? Open a ticket with the request's x-request-id (or the error body) and we'll trace it — or email support@smouter.ai.