API reference

Use MyTabulon with Codex

Connect the Codex CLI or app to a MyTabulon Coding Plan through the OpenAI-compatible Responses API. The setup is user-level, takes one key and one config block, and keeps your chosen model's full native context.

https://api.mytabulon.com/v1

Before you start

Open the Coding Plan dashboard, confirm that your chosen model is included in the current tier, then create a live mtb_live_ key with the ai.coding scope. Codex custom model providers belong in the user-level ~/.codex/config.toml file; project-local .codex/config.toml files cannot override model_provider or model_providers.

1. Set the key

Put the live key in an environment variable. Add the export to your shell profile if you want it available in every new terminal. Do not paste the key directly into config.toml or commit it to a repository.

terminal

export MYTABULON_API_KEY="mtb_live_..."

2. Add the MyTabulon provider

Codex uses the Responses protocol for custom providers. Point it at the MyTabulon v1 base URL and select any model returned by GET /models for your tier.

~/.codex/config.toml

model = "maximo-atlas-preview"
model_provider = "mytabulon"

[model_providers.mytabulon]
name = "MyTabulon Coding Plan"
base_url = "https://api.mytabulon.com/v1"
env_key = "MYTABULON_API_KEY"
wire_api = "responses"

3. Start Codex and verify

Open a new terminal so the environment variable is loaded, enter a repository, and start Codex. If you receive insufficient_scope, edit the platform key and enable ai.coding. If you receive model_not_found, choose a model included in the current Coding Plan tier.

terminal

cd /path/to/your/project
codex

# Verify the key and model catalog separately:
curl https://api.mytabulon.com/v1/models \
  -H "Authorization: Bearer $MYTABULON_API_KEY"

Switch models

Change the top-level model value in ~/.codex/config.toml, then begin a new Codex task. Examples include maximo-atlas-preview on Free, maximo-pandora-3.8-nano on Plus, openai/gpt-5.6-terra and anthropic/claude-opus-5 on Pro, and openai/gpt-5.6-sol on Max. Model-specific reasoning effort is passed through when the selected model supports it.