Agents

Agents help teams move work forward using the context already stored in Workbase. They are most useful when they are connected to specific resources, have a clear goal, and produce output that a teammate can review.

What agents can use

Agents can work with workspace context such as records, conversations, files, notes, lists, tasks, apps, skills, MCP tools, and prior agent runs. The exact context available depends on the agent, the workspace resources it is connected to, and the permissions available in the workspace.

Agent surfaces

Workspace agent

The workspace agent is the general assistant surface. Use it for questions, analysis, and guided work across the workspace.

Research workflows

Research workflows are designed for source-backed enrichment. They should prefer reliable answers, use public and workspace context where available, and return structured results instead of guessing.

Custom agents

Custom agents are configured around a repeatable job. Common examples include inbound qualification, pipeline follow-up, account research, and deliverability review.

Good agent setup

1. Define one job

   Write the agent goal as one repeatable workflow, not a broad job description.

2. Connect real context

   Attach the resources the agent should use, such as an inbox channel, object, record list, app, integration, or MCP tool.

3. Decide the handoff

   Choose what the agent should produce: a summary, a draft, a task, a record update, or a recommendation.

4. Review runs

   Check run history, status, tool calls, credit usage, and failed runs before relying on the workflow.

What to document for each agent

• What triggers the agent.
• Which resources it can read.
• Which actions it can take.
• Whether it needs human approval.
• Where output appears.
• How failed runs are reviewed.

Next guides