/ agents / codex

Codex CLI

Point OpenAI's Codex CLI at base/mnt.ai via config.toml or env vars.

OpenAI’s Codex CLI is configured by ~/.codex/config.toml and speaks the OpenAI Chat Completions protocol — which the router serves unchanged.

Option 1 — config.toml model provider (recommended)

Define base/mnt.ai as a model provider and set it as the default profile:

~/.codex/config.toml

# base/mnt.ai as an OpenAI-compatible provider
[model_providers.basemnt]
name = "base/mnt.ai"
base_url = "https://api.basemnt.ai/v1"
env_key = "SK_KEY"
wire_api = "chat"

[profiles.basemnt]
model_provider = "basemnt"
model = "z-ai/glm-5.2"

# Make it the default
model_provider = "basemnt"
model = "z-ai/glm-5.2"

env_key = "SK_KEY" tells Codex to read the key from that environment variable — the key never lives in the config file:

~/.zshrc

export SK_KEY=sk-your-key

Then run it:

run.sh

codex "refactor the auth middleware to use zod"
# or explicitly with the profile:
codex --profile basemnt "refactor the auth middleware"

wire_api

Use wire_api = "chat" for the Chat Completions surface. Codex also supports a "responses" wire API; the router does not serve the Responses protocol, so keep it on "chat".

Option 2 — env vars

Codex honors OPENAI_BASE_URL / OPENAI_API_KEY when no provider override exists:

env-vars.sh

export OPENAI_BASE_URL=https://api.basemnt.ai/v1
export OPENAI_API_KEY=sk-your-key

Option 3 — bash function

~/.zshrc

codex() {
  OPENAI_BASE_URL=https://api.basemnt.ai/v1 \
  OPENAI_API_KEY="$SK_KEY" \
  command codex "$@"
}

Verify

Run codex in any repo and issue a prompt. Requests appear in your usage dashboard. A 401 means a bad key; a 400 on /v1/chat/completions usually means the model ID is not in the catalog — set model in the profile to a valid ID.