Which Codex clients work with Model.sale?

The guide covers Codex CLI, Codex App and the VS Code Codex extension through the Responses protocol. Availability depends on the current public model catalog.

Where should I put my API key?

Keep the key in a local environment file or secret manager. Do not put it in a URL, shell command argument, repository, screenshot or browser analytics.

How do I verify the setup?

Call GET /v1/models and then send a small Responses request. Check the x-model-sale-request-id header and the Usage page for terminal usage and the exact charge.

Codex setup · updated 2026-08-26

Connect Codex to Model.sale.

Use one prepaid Model.sale key in Codex CLI, Codex App or VS Code. The connection uses the Responses protocol, keeps the secret local and charges the same wallet as the rest of your API usage.

Create an account Open interactive guides Read API docs

01 · Prepare

Account and balanceCreate a key in the dashboard and add at least $5. Funds are reserved before a request and settled from reported usage.

02 · Configure

Responses endpointUse https://api.model.sale/v1 as the base URL and select a model returned by /v1/models.

03 · Verify

First requestSend a tiny prompt, confirm the request ID, then inspect tokens and the exact charge in Usage.

Before you start

Sign in at model.sale and create an API key under API Keys. The full secret is shown once.
Deposit funds in Billing, or use the one-time Telegram test credit if your account is eligible.
Open GET https://api.model.sale/v1/models and choose a model whose status is live. Do not guess a model ID or silently substitute another model.
Have the Codex client installed, then choose either the environment-variable method or the TOML configuration below.

Security rule: never paste a real key into a URL, installer argument, Git repository, issue, screenshot or shared chat. If it is exposed, revoke it immediately and create a replacement.

Option A — local environment variable

This is the simplest setup for Codex CLI and works well for local development. The value is read by the client at runtime rather than being embedded in the command.

mkdir -p "$HOME/.config/model-sale"
umask 077
read -rsp "Model.sale API key: " MODEL_SALE_API_KEY; echo
printf 'export MODEL_SALE_API_KEY=%q\n' "$MODEL_SALE_API_KEY" > "$HOME/.config/model-sale/env"
chmod 600 "$HOME/.config/model-sale/env"
unset MODEL_SALE_API_KEY
source "$HOME/.config/model-sale/env"

Add the final source line to ~/.zshrc or ~/.bashrc if you want it loaded in new shells. Restart Codex after changing the environment.

Option B — Codex custom provider

Codex supports a custom provider in ~/.codex/config.toml. Keep the key outside this file and reference it through env_key.

[model_providers.model_sale]
name = "Model.sale"
base_url = "https://api.model.sale/v1"
env_key = "MODEL_SALE_API_KEY"
wire_api = "responses"

If you already have a config, back it up before editing. Select model_sale as the provider in the client and choose a live model from the catalog. The exact UI label can differ between Codex releases.

Codex App and VS Code

Open the provider or custom endpoint settings in the app or extension.
Set the base URL to https://api.model.sale/v1.
Choose Responses as the wire protocol when the client asks for one.
Set the API-key environment variable to MODEL_SALE_API_KEY; do not paste the secret into a project file.
Restart the app or reload the VS Code window, then select a live model.

Model.sale exposes the standard Responses path at /v1/responses. The client must preserve the requested model ID and should show an error if that model is unavailable.

Verify with a minimal request

Run this from the same shell in which Codex will run. The example uses a small prompt to avoid spending more than necessary.

curl https://api.model.sale/v1/responses \
  -H "Authorization: Bearer $MODEL_SALE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","input":"Reply exactly: OK"}'

Check	Expected result
Status	`200` with a Responses JSON object.
Request ID	`x-model-sale-request-id` is present. Keep it when contacting support.
Usage	Input/output usage appears in the response or Usage page and the reservation is settled.
Balance	The wallet decreases by the displayed charge; unused reservation is released.

Streaming

For streaming clients, set stream: true. Events are flushed as they arrive and the terminal usage event is needed for exact settlement.

Reasoning

Only request reasoning effort values shown by the current model capability. Unsupported values can be rejected before dispatch.

Limits

API-key RPM, TPM, concurrency and spend caps still apply. A request is never switched to a cheaper model to fit a limit.

Troubleshooting

Symptom	What to check
`401` unauthorized	Confirm the variable is loaded in the client process, the key starts with `ms_live_` and it was not revoked. Never print the full key to logs.
`402` or insufficient balance	Top up the wallet and check available versus temporarily reserved balance. The initial reservation is released after terminal usage.
`404` model not found	Refresh `/v1/models` and copy the exact live model ID. Do not append provider-specific aliases.
`429` rate or spend limit	Wait for the limit window, reduce parallel requests or update the key limits in the dashboard.
No model in the client	Restart Codex, verify `wire_api = "responses"`, check the base URL has no duplicate `/v1`, and test the curl request first.
Stream stops early	Retry once, record the request ID and check Usage. A disconnect is not a reason to resend repeatedly; duplicate work can be charged separately.

Rollback and key hygiene

To roll back, restore your saved ~/.codex/config.toml, remove ~/.config/model-sale/env, and remove the source line from your shell profile. Revoke the key from API Keys if the machine was shared or the secret may have been copied.

Model.sale stores request metadata needed for billing and operations, not your prompt or response body. Keep your own client logs equally careful: disable verbose request logging when it could include authorization headers or model content.

Manage API keys Open Usage Contact support

FAQ

Is this an OpenAI API key?

No. Use a Model.sale key and the Model.sale base URL. The Responses wire format is compatible with Codex clients, but the credential belongs to your Model.sale account.

Can I use any model name?

Use only IDs returned by /v1/models. The catalog changes as validation and availability change; unavailable models are not accepted for customer traffic.

Where can I see what I spent?

Open Usage after the request. It shows the endpoint, model, status, token usage, charge and request ID without storing prompt content.