Skip to main content
Custom Agents are Forge’s configurable background workers. They can gather information, produce reports, prepare operational outputs, and use allowed connected tools on demand or on a schedule. In the product:
  • Agent Library is where you find and configure Custom Agents.
  • Agent Runs is where you inspect their execution, output, recovery, and follow-up work.

​
What belongs in a Custom Agent

Choose work that is:
  • repeated often enough to justify a maintained workflow;
  • driven by clear inputs and scope;
  • observable through a useful output;
  • safe to run with explicit tool boundaries;
  • reviewable by an owner;
  • valuable even when it correctly returns “no action needed.”
Examples include account health summaries, recurring operational checks, renewal or usage review, exception reports, and preparation work for a human process.

​
The agent lifecycle

1

Define the outcome

Name the decision or artifact the agent should produce and who uses it.
2

Configure the work

Choose instructions, model, tools, skills, files, scope, schedule, output, and optional judges.
3

Dry run

Test representative targets without write actions.
4

Review the evidence

Check sources, gaps, output structure, and judge results.
5

Activate

Make the agent active only after its owner accepts the behavior and tool boundary.
6

Operate

Monitor runs, recover failures, follow up in Chat, and version meaningful changes.
A scheduled agent has no person present to answer an approval request. Tool access for Custom Agents is controlled through Availability, roles, scope, and the agent’s selected tools—not interactive approval prompts.
Start with Create a Custom Agent.