FAQ

Does it cost anything? agentctl itself is free. What you pay is what your provider charges. With a free-tier key the report says $0.00 (free-tier model). A list price litellm attaches to a free call is labelled as such, and not counted as spend.

Does it send my code anywhere? Only to the model provider you configured, which is what any coding agent does. Your keys stay in ~/.agentctl/keys.env and are never printed or logged. The ledger, the run index and recorded conversations stay on your machine.

Which model should I use? Whatever agentctl init recorded is a model that answered. Stronger models write better code. agentctl does not change that, and --accept tells you whether a run worked.

Can I use more than one key at the same provider? agentctl accepts them, but whether a second key there is a second quota is unverified: OpenRouter's own documentation says extra accounts do not change rate limits. Pooling several free accounts may also break a provider's terms. A key at a second provider is the reliable way to survive a cap.

Is it a sandbox? No. Commands run on your machine. agentctl asks before dangerous ones and stops actions from repeating across a crash, but an allowed command does whatever it does. See concepts.md.

What happens if I run two terminals on the same task? The second agentctl resume refuses and names the process that holds the conversation. Two drivers of one conversation is the thing agentctl exists to prevent.

Why did my test re-run not get blocked? Because the agent had already seen the first result, re-running is its decision. That is the normal edit, test, re-test loop. agentctl only stops an action from repeating when a crash, not the agent, would repeat it.

Where is everything kept?

How do I check that it works on my machine? agentctl demo in a minute. python verify.py from a clone runs the whole correctness suite, in about five minutes, at no cost.