Connect Codex CLI

Updated 2026-09-28

OpenAI's terminal coding agent, connected over the Responses API.

Use cases

Good for letting an agent read your project, edit files and run tests from the terminal. Codex CLI uses the Responses API (wire_api = "responses") and reads the key from an environment variable instead of the config file.

Tips

A coding agent sends many long requests per task, so pick a long-context model with tool calling and set a daily spend limit on the key.

Reasoning effort

Codex CLI sends model_reasoning_effort from config.toml as reasoning.effort. The platform follows the levels the serving line accepts: a level the upstream rejects becomes the next higher accepted level, or its highest one. If the model cannot turn reasoning off, asking for it returns 422 model.reasoning_off_unsupported. Each model's table is under Reasoning effort on its page in the catalog; rewritten calls show the rewrite in usage records.

If it cannot connect

1. Set base_url to https://omnimodel.me/v1, without /responses.
2. The key starts with key_ and is not paused or revoked; auth.invalid_key means the key cannot be used.
3. The model ID matches the catalog and is in the key's allowed models.
4. billing.insufficient_credit means the balance is too low; see common errors for more.

Setup

1. Install: npm install -g @openai/codex.
2. Export the key in your shell profile: export A2AGENT_API_KEY=.
3. Put the block below at the top of ~/.codex/config.toml, then run codex in your project.

Settings

• API host：https://omnimodel.me/v1
• API key：YOUR_API_KEY
• Model：MODEL_ID

Compatibility testing for Codex CLI is still in progress; if something breaks, open a ticket with the request ID.

model = "MODEL_ID"
model_provider = "a2agent"

[model_providers.a2agent]
name = "A2Agent"
base_url = "https://omnimodel.me/v1"
env_key = "A2AGENT_API_KEY"
wire_api = "responses"