Agents FAQ
Links on this page to features marked Preview describe designs that are not available yet.
When should I use the Agents API instead of chat completions with tools?
Use agents when you want Syntic to keep the conversation, run tools in a sandbox, wait for approvals, or run on a schedule. If you already run your own loop and store history, chat completions with tools is simpler.
Where do agent tools run?
In a Syntic-hosted environment, on your self-hosted runners, or in your own code for function tools with executor: client.
How do I stop an agent from doing something dangerous?
Restrict its tools and put risky ones behind approvals. Prompts help, but permissions are the control. Test with safety evaluations.
How do I cap what a session can cost?
Set session_cost_budget_usd and max_turns on the agent or per session; see Budgets & limits.
Can an agent remember a customer between sessions?
Yes, with memory stores scoped per customer, and user profiles for stable facts.
How do I see what a sub-agent did?
List the session’s threads and read the delegate thread’s events.
Was this page helpful?