Agents won't start or respond
Work down this list — it’s ordered by how often each cause is the culprit: the CLI isn’t installed, the CLI isn’t signed in, the key isn’t set (or a stale window is running), or the task hit an error the transcript explains.
The window says the CLI isn’t available
Section titled “The window says the CLI isn’t available”Cause: the agent’s binary isn’t installed or isn’t on your PATH.
- Open Settings → Agents — the detection list shows every brand with found/not-found and an Install hint.
- Install the CLI (each brand’s official installer), then click Scan again.
- Open a new agent window of that brand.
The agent replies with a login/authentication error
Section titled “The agent replies with a login/authentication error”Cause: the CLI isn’t signed in and no API key is set.
- Easiest: run the CLI once in a terminal outside Mastery HQ and complete its login flow.
- Or set its API key in Settings → Agents → Agent API keys (.env), click Recheck keys, and open a fresh agent window. (Key guide)
- The transcript shows the exact login instruction for the brand when this happens — follow it verbatim.
I added a key but the agent still fails
Section titled “I added a key but the agent still fails”Cause: running windows keep their original environment.
- Close the agent window and open a new one — new windows pick up the current
.env. - Check for a project-level
.envin the workspace folder overriding your global keys.
The run starts, then exits immediately with a code
Section titled “The run starts, then exits immediately with a code”Cause: the task itself errored — the CLI’s message is in the transcript just above Process exited.
- Read the last lines before the exit; they name the problem (rate limit, invalid flag, network).
- Retry once; if it persists, run the same prompt in that CLI directly in a terminal to see its full output.
Nothing happens at all when I press Send
Section titled “Nothing happens at all when I press Send”- Confirm you’re in the desktop app (agents require it — the browser preview can’t spawn processes).
- Check who’s working — if the agent is mid-run, new input waits its turn; Stop first to interrupt.
Still stuck?
Section titled “Still stuck?”Export the transcript (Export button) and send it via Feedback with the agent brand and what you expected.