# Codex

> Add Tchavi as a model provider in ~/.codex/config.toml and run Codex on mobile-money credits.

- Canonical: https://tchavi.com/en/docs/coding-agents/codex

---


Part of the [coding agents](/docs/coding-agents) guides.

Codex describes providers as named blocks in `~/.codex/config.toml`. Tchavi is one block and one environment variable.

## Configure it

```toml
model = "gpt-5.4"
model_provider = "tchavi"

[model_providers.tchavi]
name = "Tchavi"
base_url = "https://tchavi.com/api/v1"
env_key = "TCHAVI_API_KEY"
wire_api = "chat"
```

Then:

```bash
export TCHAVI_API_KEY=sk-tch-your-key
codex
```

`env_key` names the variable Codex reads the key from — the key itself never enters the file, so `config.toml` stays safe to sync between machines.

## wire_api

`wire_api = "chat"` selects `/v1/chat/completions`, the format every model in the catalogue supports.

<Callout type="note" title="Which wire_api to pick">
  Tchavi also serves `/v1/responses`, so `wire_api = "responses"` works for OpenAI models. Use
  `"chat"` unless you specifically need a Responses-only feature: it is the format that also covers
  the Claude, DeepSeek, Gemini and Qwen models, so switching model becomes a one-line change.
</Callout>

## Switching models

Change `model` to any Tchavi model id — `claude-sonnet-4-6`, `deepseek-v4-pro`, `gpt-5.4`. To keep several providers configured at once, add more `[model_providers.*]` blocks and point `model_provider` at whichever you want:

```toml
[model_providers.tchavi-cheap]
name = "Tchavi (budget)"
base_url = "https://tchavi.com/api/v1"
env_key = "TCHAVI_API_KEY"
wire_api = "chat"
```

Both blocks can share one `env_key`, since it is the same account.

---

Next: [Cursor](/docs/coding-agents/cursor) · Back to [coding agents](/docs/coding-agents)