Agent Settings
Quick reference for everything under an agent's Settings tab. For the guided version, see Creating & Configuring Agents.
Most tabs show an Unsaved changes bar when you edit something — changes take effect when you hit Save. The one exception is Relays, where target grants save immediately.
General
| Setting | Description |
|---|---|
| Name | Short identifier like code-reviewer. Unique among siblings; other agents use it when delegating. |
| System Prompt | The agent's core instructions. Injected into every run, never compacted — the most important field on the whole page. |
| Hierarchy | Standalone or child of another agent. Determines who can delegate to it and which /shared volume it mounts. Use the tree picker to re-parent. |
Model
| Setting | Description |
|---|---|
| Model | The AI model, grouped by provider, with context window and pricing shown. See Models. IDs starting with subscription/ run on a Claude or ChatGPT subscription of your own — see Subscriptions. |
| Reasoning Effort | How hard the model thinks before responding. Only shown for models that support it. Higher = more thorough, slower, pricier. |
| Apply to all children recursively | Push the model choice down the whole subtree in one click. Shown only when the agent has children. |
Tools
A checkbox per tool, plus quick-select buttons: All, Read-only (only tools that don't modify state — good for review agents), and None. See Tools for what each one does.
Schedule
Timer-driven runs. Full guide: Scheduled & Autonomous Agents.
| Setting | Description |
|---|---|
| Schedule | Presets from every 5 minutes to monthly, or Custom — natural language or a raw cron expression. |
| If Already Running | What happens when the schedule fires mid-run: Skip (default), Queue, or Kill & Restart. |
| Run Mode | Start a new run each time (default) or Continue the last run, which keeps context across triggers. |
| Next Run | Live countdown to the next trigger. Click it to fire a scheduled run immediately. |
Container
Only visible when your deployment exposes infrastructure overrides — if you don't see it, the defaults are in use. Lets you set a custom Docker image, CPU/memory/process limits, and bind mounts.
Relays
Grants for remote execution targets (both relay and SSH targets, despite the tab name). Per target you toggle access, choose allowed operations (Shell, Read, Write, Sync), and optionally set path-based rules for file operations. Grants save immediately.
Note that path policy constrains file operations only — it does not sandbox Shell. See Connecting Your Own Machines.
Advanced
| Setting | Default | Description |
|---|---|---|
| Max Turns | Unlimited | Cap on turns per run — a useful safety net for scheduled agents. 0 means unlimited. |
| Timeout | 5 minutes | Max wall-clock time for a run. |
| Disable Agent | Off | The agent keeps its configuration but won't start runs or accept scheduled triggers. |
| Timestamp All User Turns | Off | Prepends date/time to every user message, for agents that need time awareness. |
| Full System Access | Off | Grants the account-wide Admin tool. Requires confirmation. See System access. |
At the bottom, the Danger Zone holds Delete Agent — this permanently removes the agent, its descendants, their containers, and run history.