OpenAI

gpt-5.x. mantis handles this API's quirks for you — max_completion_tokens vs max_tokens and locked temperature are auto-negotiated.

Get an API key

1. Sign in at platform.openai.com
2. Open the API keys page
3. Click 'Create new secret key'
4. Name it and copy the key (shown once)
5. Add funds under Settings → Billing

Create a key ↗ · Pricing ↗

No free tier; pay-as-you-go needs a prepaid funded balance.

Enable

export OPENAI_API_KEY=...        # shell profile — survives forever

or in the terminal — validates the key live before saving:

/enable openai

or pick any locked 🔒 OpenAI model in /models and paste the key inline.

Models

Starter menu (once enabled, /models fetches the provider's full live list):

• gpt-5.4
• gpt-5.4-mini
• gpt-5.4-nano
• gpt-5.4-pro

Switch anytime — /model gpt-5.4 fuzzy-matches; context carries over.

SDK

from mantis_agent import MantisAgentOptions
options = MantisAgentOptions(
    model="gpt-5.4",
    backend="https://api.openai.com/v1",     # key read from $OPENAI_API_KEY
)

Notes

• gpt-5.x rejects max_tokens and non-default temperature; mantis retries with the right parameters automatically.
• Realtime/audio/codex models are filtered out of /models — only chat models show.