Agent Studio User Guide
Design, test, publish, deploy, and monitor AI agents and multi-step workflows with tools, knowledge, schemas, runs, and endpoints.
Who This Guide Is For
- Agent builders
- AI application teams
- Automation engineers
Where To Go
| Page |
Use It For |
/agentstudio |
Agent Studio overview. |
/agentstudio/agents |
Agent catalog. |
/agentstudio/workflows |
Workflow builder and workflow list. |
/agentstudio/tools |
Tool registry. |
/agentstudio/knowledge |
Knowledge sources for agents. |
/agentstudio/schemas |
Input/output schemas. |
/agentstudio/runs |
Execution history. |
/agentstudio/endpoints |
Deployed agent endpoints. |
/agentstudio/templates |
Reusable agent and workflow templates. |
/agentstudio/analytics |
Usage and performance analytics. |
Core Concepts
| Concept |
Meaning |
| Agent |
A configured AI worker with model, instructions, tools, memory, and policies. |
| Workflow |
A graph or sequence of steps that coordinates agents, tools, and decisions. |
| Tool |
An action the agent can call, such as API access, retrieval, or platform operation. |
| Run |
One execution with inputs, trace, outputs, and status. |
| Endpoint |
A deployed callable interface for an agent or workflow. |
Common Workflows
Build and test an agent
- Create an agent.
- Choose model, system instructions, and tools.
- Attach knowledge if needed.
- Define schemas and safety settings.
- Run test conversations or workflow runs.
- Review traces and outputs.
- Publish or deploy when quality is acceptable.
Debug a failed run
- Open Runs.
- Select the failed run.
- Review step trace, tool calls, inputs, outputs, and errors.
- Update prompt, schema, tool config, or knowledge source.
- Run again with the same test case.
Best Practices
- Use explicit schemas for production workflows.
- Keep tool permissions scoped to the agent's actual job.
- Review run traces before publishing changes.
- Use templates for repeatable agent patterns.