Lifecycle
Start, stop, restart, and delete an agent from the console.
Each agent is a running service you control from its console page. Start it to bring it online, Stop it to take it offline (and pause credit usage), Restart it to pick up a saved change, or Delete it entirely.
This is one of the core differences from calling models yourself: an agent is a long-running service with a lifecycle, not a one-off request. NevTan supervises it so you don't have to.
The lifecycle
Status and controls
The agent list shows each agent's current status — running, paused, or stopped. Open an agent to find its lifecycle buttons:
| Button | What it does |
|---|---|
| Start | Brings a stopped agent back online. |
| Stop | Takes the agent offline and pauses its credit usage. |
| Restart | Restarts the agent so it picks up a saved configuration change. |
| Delete | Permanently removes the agent and its memory. |
What each action preserves or resets
It matters which of these you reach for, because they don't all touch the same things. Configuration — channels, credentials, provider, model, and the memory/web-search toggles — is saved on the agent record itself, not on the running process, so nothing here wipes it out short of deleting the agent.
- Stop → Start — the agent goes fully offline and stops using credits, then comes back exactly as it was: same channels, same credentials, same persistent memory. Nothing is reset. Use this for "I don't need this running right now" without losing anything.
- Restart — reloads the agent so it picks up saved changes (a newly connected channel, a rotated key, a different model). Persistent memory is not cleared by a restart — it's tied to the agent, not to a single run of the process.
- Delete — the only irreversible one. It removes the agent and its persistent memory permanently; there's no configuration left to restart into afterward.