OpenAI¶

Missy supports two OpenAI-compatible providers: the standard OpenAI provider (Chat Completions API) and the Codex provider (ChatGPT backend API with OAuth).

OpenAI provider¶

Setup¶

Environment variable (recommended)Config file

export OPENAI_API_KEY="sk-..."

providers:
  openai:
    name: openai
    model: "gpt-4o"
    enabled: true

providers:
  openai:
    name: openai
    model: "gpt-4o"
    api_key: "sk-..."
    enabled: true

Verify¶

missy providers
missy ask "Hello" --provider openai

OpenAI-compatible endpoints¶

The OpenAI provider works with any API that implements the Chat Completions format. Set base_url to target alternative services:

providers:
  groq:
    name: openai
    model: "llama-3.1-70b-versatile"
    api_key: "gsk_..."
    base_url: "https://api.groq.com/openai/v1"
    enabled: true
  together:
    name: openai
    model: "meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo"
    api_key: "..."
    base_url: "https://api.together.xyz/v1"
    enabled: true

Network policy

Provider base_url hosts are automatically added to provider_allowed_hosts, so you usually do not need to update the network policy manually.

Models¶

Model	Config value	Notes
GPT-4o	`gpt-4o`	Default -- multimodal, fast
GPT-4o mini	`gpt-4o-mini`	Cheaper, good for simple tasks
GPT-4 Turbo	`gpt-4-turbo`	Older, larger context

Configuration reference¶

Field	Type	Default	Description
`name`	string	`"openai"`	Must be `"openai"`
`model`	string	`"gpt-4o"`	Model identifier
`api_key`	string	null	API key (falls back to `OPENAI_API_KEY` env var)
`base_url`	string	null	Override API endpoint for compatible services
`timeout`	int	`30`	Request timeout in seconds
`enabled`	bool	`true`	Enable/disable

Codex provider¶

The Codex provider uses ChatGPT's backend API with OAuth authentication. This is useful if you have a ChatGPT subscription and want to use it with Missy.

Setup via wizard¶

The easiest way to configure the Codex provider is through the setup wizard:

missy setup

Select OpenAI (OAuth/Codex) when prompted. The wizard handles the OAuth flow and stores the token securely at ~/.missy/secrets/openai-oauth.json.

Manual configuration¶

providers:
  openai-codex:
    name: openai-codex
    model: "gpt-4o"
    timeout: 60
    enabled: true

The OAuth token is loaded automatically from ~/.missy/secrets/openai-oauth.json and refreshed when near expiry.

How it differs from the standard OpenAI provider¶

Aspect	OpenAI provider	Codex provider
Endpoint	`api.openai.com`	`chatgpt.com/backend-api`
Auth	API key	OAuth bearer token
Request format	Chat Completions	Responses API
Extra headers	None	`chatgpt-account-id` from JWT

Dependencies¶

Both providers require the openai Python package:

pip install -e "."

The Codex provider additionally uses httpx (bundled with Missy's dependencies).