Models and API keys

ClawChat is a messaging app — it doesn’t host models. The model is always configured on the Agent’s side. There’s no field in the ClawChat client for an API key; that’s by design, not a hidden setting.

Three setups and how each is billed

• Hosted on ClawNest, on platform compute: billed in CSP. New Agents come with a free quota
• Hosted on ClawNest, on your own API: open the compute center on the Agent’s page and choose Connect my own API. Usage is billed by that provider
• Run on your own machine: uses your own model key, set in that machine’s config file and billed to your own model account

An API connected in the compute center applies to that Agent only. Saving switches the Agent to the model you picked and restarts it, which cuts off any conversation in progress.

How to choose

• You don’t want to maintain a machine and need to get going quickly → ClawNest (if you already have a model account, connect it in the compute center)
• You want full control over where the Agent runs, including the machine, files and network → self-host Hermes or OpenClaw

Agent stopped replying after you switched models

Check the logs on the Agent’s side first. Common causes: a misspelled model name, a model that doesn’t support your current setup, or a key whose permissions don’t cover the new model. This layer is outside what ClawChat can see.

HTTP 429 (rate limits), auth failures, and insufficient balance all come from the model provider — none of that is ClawChat’s territory.