Model settings & token plans

Libra OS routes work across three model tiers. Each tier is just a model id behind an OpenAI-compatible endpoint, so configuration is a key, a base URL, and your model ids — nothing else.

Tier	What it does	Setting
Brain	Plans the task and makes the hard calls	LIBRA_OS_BRAIN_MODEL
Skill	Runs the cheap, high-volume agent steps	LIBRA_OS_SKILL_MODEL
Answer / synthesis	Writes the final grounded answer	OPENAI_MODEL

The gateway itself is set with OPENAI_API_KEY and OPENAI_API_BASE. The canonical prefix is LIBRA_OS_*; the legacy NOVA_OS_* names are bridged and keep working forever, so renaming existing deployments is optional. Any OpenAI-compatible provider works — bring your own keys, point the tiers at local models, or mix per tier.

Using your token plan

Token plans (from $249/month — see pricing) include an enterprise-model usage allowance. To draw from the allowance instead of paying per token:

1. Use the plan key. Your subscription page provisions a dedicated API key when the plan activates. The allowance only applies to that key — usage on any other key, even for the same models, is billed pay-as-you-go.

2. Set the base URL to the bare gateway host, without /v1:

OPENAI_API_KEY=sk-<your plan key>
OPENAI_API_BASE=https://api.meganova.ai
LIBRA_OS_BRAIN_MODEL=anthropic/claude-opus-5-Ent
LIBRA_OS_SKILL_MODEL=gemini/gemini-2.5-flash-lite-Ent
OPENAI_MODEL=gemini/gemini-2.5-pro-Ent

3. Use only covered model ids, in the full provider/...-Ent form. Bare names are rejected, and any non-covered model called with the plan key is billed pay-as-you-go rather than drawing from the allowance. The current covered list is published on your plan's subscription page.

Audio is covered too: transcription with Systran/faster-whisper-large-v3 at POST /v1/audio/transcriptions on the same gateway draws from the same allowance (billed per second).

Hot reload — change models without a restart

Settings persist and reload at runtime via the settings API:

curl -X PUT http://localhost:8000/api/config/settings \
  -H 'Content-Type: application/json' \
  -d '{
    "openai_api_key": "sk-<your plan key>",
    "openai_api_base": "https://api.meganova.ai",
    "synthesis_model": "gemini/gemini-2.5-pro-Ent"
  }'

This is also how key rotation works: regenerate the key on the subscription page, then hot-reload the new value — the old key stops working immediately.

Troubleshooting

Symptom	Likely cause
Requests billed pay-as-you-go instead of the allowance	Wrong key, or a non-covered model id
404 on model calls	Missing provider/ prefix, or base URL includes /v1
Requests hard-stop mid-month	Allowance exhausted and extra usage not enabled — enable it (with a monthly cap) on the subscription page

The allowance resets at the start of each billing period. Overage is opt-in and capped — it is never charged automatically.