Token Supply

Setup guide

Use Codex with a custom API endpoint

Yes. Codex works with Token Supply. Keep the Codex CLI, app, or VS Code extension you already use. Point the provider at Token Supply, set your API key, and keep the Responses API.

~/.codex/config.toml · ~/.codex/auth.json

Create account Open docs

Four files and a restart

  1. Create an API key

    Register, add credit, and create a key in API Keys. A new key works for every client. Copy ts_live_... once.

  2. Edit config.toml

    Set model_provider, model, reasoning effort, and a Token Supply provider block.

  3. Add auth.json

    Store the Token Supply key as OPENAI_API_KEY.

  4. Restart Codex

    Select the configured model. A short test request should appear in Usage.

Who this is for

Codex CLI, Codex App, and VS Code extension users who already run an OpenAI-plan key and want the Responses API left intact.

What you need

Use your Token Supply API key. A new key works for every client. The model must be one Codex accepts. Older keys still only serve the plan they were created for.

This setup works for Codex CLI, Codex App, and the Codex VS Code extension.

  • Base URL: https://api.token-supply.com
  • wire_api must be responses. Do not use chat.
  • disable_response_storage and store: false. Codex should not persist Responses on the provider.

Edit config.toml

Open ~/.codex/config.toml. On Windows, open %USERPROFILE%\.codex\config.toml.

Keep the normal Codex provider setup and change the values to Token Supply. Do not put the provider block in a project .codex/config.toml; Codex ignores provider settings there.

~/.codex/config.toml
model_provider = "token_supply"
model = "gpt-5.6-sol"
model_reasoning_effort = "high"
model_reasoning_summary = "auto"
model_context_window = 1050000
disable_response_storage = true
supports_websockets = false

[model_providers.token_supply]
name = "Token Supply"
base_url = "https://api.token-supply.com"
wire_api = "responses"
requires_openai_auth = true
supports_websockets = false

[features]
goals = true

Why this works: wire_api stays responses, so Codex keeps the API it already calls.

  • On Windows, use %USERPROFILE%\.codex for both files.
  • wire_api must be responses. Do not use chat.
  • model_reasoning_effort accepts minimal, low, medium, high, or xhigh for the example model. gpt-6.1-sol accepts low, medium, high, xhigh, or max, and does not accept none or minimal.
  • GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, and GPT-5.5 use a 1,050,000 token context window.
  • gpt-6.1-terra is also an OpenAI-plan id. The example model stays gpt-5.6-sol. Its rates are not the rates for other models.

Add the key

Open ~/.codex/auth.json and use your Token Supply key as the OpenAI key.

~/.codex/auth.json
{
  "OPENAI_API_KEY": "ts_live_..."
}

Why this works: The key lives in auth.json, separate from the provider block.

Verify

Restart Codex and select gpt-5.6-sol, or another OpenAI id such as gpt-6.1-sol, gpt-6-sol, gpt-6-luna, or gpt-6.1-terra, on that key's plan. Send a short prompt, then open Usage in Token Supply.

If you want a lower listed rate on the same Codex workflow, compare models in the savings calculator before you change the config.

Open the savings calculator · Reduce coding-agent API costs

Details

Does Codex work with Token Supply?
Yes. Codex CLI, Codex App, and VS Code extension users who already run an OpenAI-plan key and want the Responses API left intact.
Why responses instead of chat?
Codex uses the Responses API. wire_api must be responses or the client will call the wrong path.
Can I keep my current Codex model name?
Use a model included in the Token Supply plan on that key. The name in the snippet is an example.
Where do I put the key on Windows?
Both config.toml and auth.json live under %USERPROFILE%\.codex.

Related