Skip to content
Fast Inference
DOCS / Codex
Integration

Codex

Route Codex through Inference.net's OpenAI-compatible Responses API.

fast codex on adds an inference-net provider to ~/.codex/config.toml (Responses API at https://api.inference.net/v1) and makes it the default. The lines it manages are wrapped in # inference-fast comments.

Connect Codex

  1. Install

    bash
    npm install -g @inference/fast

  2. Activate

    bash
    fast codex on

    The first run signs you in through the browser and creates an API key for this machine. The command backs up Codex's config before editing it.

  3. Restart Codex

    Restart Codex and send a prompt; the request appears under Usage.

Choose a model

Codex uses kimi-k3-fast unless you pick another ID from the model catalog:

bash
fast codex on --model kimi-k3-fast

Run on again to change the default. Restart Codex afterward.

Switch models in Codex

on writes every model from the model catalog into a Codex model catalog (~/.codex/inference-fast-models.json), so /model lists them all with the same system prompt and tools Codex uses for its own models. A /model choice lasts for the session; run fast codex on --model <id> to change the default. Codex must be installed when you run on, otherwise the picker is not configured.

Check status

bash
fast codex status

Shows the connection, model, and config paths.

Disconnect

bash
fast codex off

Restores config.toml to the backup taken by on, so edits made after connecting are lost. Restart Codex afterward.

Troubleshooting

  • fast codex status says off but Codex is connected — run fast codex commands with the same CODEX_HOME you used for on.
  • Requests return 401 — run fast status, then fast codex on.
  • Model rejected — pick an ID from the model catalog.