# Codex CLI

Connect → Codex CLI

# Use AxForge with Codex CLI

    Codex CLI is OpenAI's coding agent
    for the terminal. It speaks the Responses API, and AxForge serves that
    too. Add AxForge as a provider in its config file and it runs on your key.

    **Before you start**

        - An AxForge API key. Get one (a minute).

        - Node.js 18 or newer. This guide was tried with Codex CLI 0.163.0 on 2026-10-17.

## Set it up

### Install Codex

```
npm install -g @openai/codex
```

### Add AxForge to its config

Create `~/.codex/config.toml` (on Windows `%USERPROFILE%\.codex\config.toml`):

```
model = "qwen3.8-27b-nvfp4"
model_provider = "axforge"
model_context_window = 262144

[model_providers.axforge]
name = "AxForge"
base_url = "https://api.axforge.ai/v1"
env_key = "AXFORGE_API_KEY"
wire_api = "responses"
```

The file holds no key: `env_key` tells Codex to read it from the
        `AXFORGE_API_KEY` environment variable. `model_context_window` tells it that our models
        read 262,144 tokens at once, so it shortens a long session before it reaches that.

### Put your key in the environment

```
export AXFORGE_API_KEY="orx_live_your_key_here"
```

```
$env:AXFORGE_API_KEY = "orx_live_your_key_here"
```

It lasts until you close this terminal. Start Codex from the same one.

### Start it in your project

```
cd my-project
codex
```

The first time in a folder it asks whether you trust it. **Trust and continue** lets it read and change
        the files there.

### Give it a task

Ask in plain words. The line under the message box names the model it runs on.

## The settings

        | Setting | Value |  |

        | base_url | https://api.axforge.ai/v1 |  |

        | wire_api | responses |  |

        | env_key | AXFORGE_API_KEY (the variable that holds your key) |  |

        | model_context_window | 262144 |  |

        | model | qwen3.8-27b-nvfp4, glm-5.3-flash or mimo-v2.6-flash |  |

Which model for what: `qwen3.8-27b-nvfp4` is the
    everyday default and reads images, `glm-5.3-flash` thinks before it answers,
    `mimo-v2.6-flash` is the fastest. To try another one for a single run:
    `codex -m glm-5.3-flash`.

## Check that it works

    The line under the message box names your model and a task gets done. The calls show in the
    console under API Keys → Recent calls within a minute.

## When it does not work

      | What you see | What it means | Fix |  |

        | "Missing environment variable: `AXFORGE_API_KEY`" | the key is not in this terminal | set it as in step 3, in the same terminal, then start Codex again |  |

        | "Model metadata for `qwen3.8-27b-nvfp4` not found" | Codex knows only OpenAI's model names | nothing: it is a warning, and it runs |  |

        | "unexpected status 401 Unauthorized: Invalid API key. Create one at https://console.axforge.ai/keys." | the key was refused | paste it again, whole, or make a new one |  |

    Test a call says in one click whether a problem is your key,
    your balance or us.

      ← Claude Code
      Cline →



Source: https://dev.axforge.ai/docs/connect/codex/