Model keys

No key is needed to try the room. Add keys to use real models.

export COALITIONKIT_META_API_KEY=...
export COALITIONKIT_BASETEN_API_KEY=...
coalitionkit up

The host reads keys from the environment. It never prints or saves them, and they never appear in logs or errors.

Which model does what

Role	Default model	Does
Gardener	muse-wonder-1.0	Reads the goal, and names and explains plans. It never builds them: fixed rules do.
Hosted agent	muse-spark-1.3	Asks your questions, makes offers, routes what you say.
Composer	muse-spark-1.3	Lays out results from the closed catalog.
Steward audit	meta-models/muse-spark-open on Baseten	Reviews each message. It is a different model from the agents on purpose.
Vault	muse-glimmer on local Ollama	Used by a local vault.

Change a model with its setting, such as COALITIONKIT_COMPOSER_MODEL. All settings are in Environment and settings.

What a room costs

The room shows the cost it measured while it runs. In our test runs, a two or three person room cost about $0.10 to $0.25 for the host's own model calls. Yours will differ with the goal, the models and the prices. An agent you bring yourself pays for its own model.

When a call fails

Every model step has a rule-based fallback, so a failed call does not stop the room. The screen shows model or rules beside what each one made.