Codex
Codex lets you add a model provider of its own in config.toml, and that is all the Model API needs: Codex speaks the OpenAI Responses API, and the Model API answers it at POST /v1/responses. Nothing to install besides Codex itself. The model is auto, which picks a model of the catalogue for the conversation.
Requirements: Codex CLI (codex --version shows the version), a Model API key and money on the organization balance.
1. Create a key
In the developer console, open API keys, keep Model API key in the Access field and create the key. The secret is shown once. Put it into your shell profile (~/.zshrc, ~/.bashrc) so Codex can read it:
2. Add the provider
Codex reads ~/.codex/config.toml, or config.toml in the folder named by CODEX_HOME. Add the lines below. Keep the root keys (model, model_provider, model_reasoning_effort) above any [table] header of the file: in TOML a key after a header belongs to that table.
| Setting | What it does |
|---|---|
model | The model sessions start on: auto or an id from the Models page. |
model_provider | Makes the provider below the one Codex calls. |
model_reasoning_effort | The reasoning level Codex asks for: minimal, low, medium, high, xhigh or max. Codex sends none of its own for a model it does not know. |
base_url | The Model API's address. Keep /v1 at the end: Codex appends /responses itself. |
env_key | The environment variable Codex reads the key from. |
wire_api | responses, the only protocol Codex speaks. |
Restart your Codex sessions. For one session, codex --model <id> names another model.
3. Check the connection
Run in a project folder:
The answer comes back, and the call shows up in the call log of the console with the model auto picked and its cost. At the start of a session Codex warns that it has no metadata for auto and falls back to its defaults: that is expected.
How Codex works through the API
- Codex keeps no state on the server: every request carries the whole conversation with
store: false, as the Responses API expects. - For a model it does not know Codex gives the model its shell tool and its default system prompt, and edits files with
apply_patchthrough that shell. Tool calls and their results travel as Responses items. - The model's reasoning comes back as a
reasoningitem; Codex sends it back in the next request, and the model gets it again, on another host too. - Codex names the conversation in
prompt_cache_key, which keeps its calls in one cache. - Codex counts the price of a thread on its own side and does not read
usage.cost. What the balance was charged is in the call log of the console.
Undo the setup
Remove the model, model_provider and model_reasoning_effort lines and the [model_providers.samreshuuu] table from config.toml. Codex then goes back to your usual provider.
Troubleshooting
| What you see | What to check |
|---|---|
401 or a missing environment variable | SAMRESHUUU_API_KEY is not set in the shell that started Codex, or the key was revoked. |
| Codex refuses the config file | A root key such as model stands below a [table] header, or wire_api is not responses. |
model_not_found | The model id is not in the catalogue: take one from GET /v1/models or use auto. |
402 | The balance does not cover the call's hold, or the key's spend limit is reached. |
| HTTP | type | Code | message | Retry |
|---|---|---|---|---|
| 401 | authentication_error | invalid_api_key | The API key is unknown, disabled, revoked or expired. Switch a disabled key back on in the developer console; create a new one in place of a revoked or expired key. | no |
| 404 | not_found_error | model_not_found | The model does not exist. GET /v1/models lists the models this API sells. | no |
| 400 | invalid_request_error | no_auto_model | No model meets the request for auto. | no |
| 402 | billing_error | insufficient_balance | The organization balance does not cover this request (max_tokens is held in full until the answer is settled). Top up the balance or lower max_tokens. | no |
| 402 | billing_error | key_limit_exceeded | This API key has reached its spending limit. Raise the limit or use another key. | no |
If an error persists, write to support with the x-request-id of the answer.
Frequently asked questions
Why does Codex warn that model metadata for auto is not found?
Codex keeps a built-in list of OpenAI models and does not know auto, so it runs on its default settings: its own system prompt, the shell tool, file edits through apply_patch in that shell. The warning is expected and the session works.
Can Codex use wire_api = "chat"?
No. Current Codex speaks only the Responses API and refuses a config with wire_api = "chat". The Model API serves Responses at POST /v1/responses.
Where do I see what a session cost?
In the call log of the developer console (https://console.samreshuuu.ru). Codex counts the price of a thread on its own side and does not read the cost from the answer.