Custom Agents and Agent Builder

What's New

Custom agents turn the way your team handles recurring work into agents that anyone with access can run. Each agent carries its own instructions and context, such as vault files, knowledge sources, connectors, and playbooks, and starts from that same foundation every time it runs.

A custom agent works toward the outcome you describe. It adapts to the matter in front of it and asks for the inputs it needs at the start of a run or when they become relevant.

Build and Refine Agents in Agent Builder

The new Agent Builder is Harvey’s text-based experience for creating custom agents, enabling you to type or speak instructions for the agent you want to create. You describe the result the agent should produce, and Harvey drafts the agent’s instructions with you, asking follow-up questions to sharpen its behavior. If you already have instructions written, you can also import an existing skill file (.md, .skill, or .zip).

The drafted instructions appear in an editor alongside your thread. You can edit the agent’s instructions directly, ask Harvey to make changes, and attach the context the agent should draw on.

When the agent is ready to try, run a test and tell Harvey what to adjust. You can review each round of edits as a redline before you save.

Find, Run, and Schedule Agents

The new Agents page brings together the agents you’ve built, agents shared with you, and agents built by Harvey. You can run an agent from its details page or call it into any thread using an @ tag. When your request matches an agent available to you, Harvey may also suggest running it. For work that repeats on a regular cadence, you can schedule an agent to run on a recurring basis.

Share and Manage Agents

Saved agents can be shared with individuals, groups, spaces, or your entire workspace. Admins control who can build agents, can manage every agent in the workspace from Command Center, and set when sharing an agent requires approval.

Why It Matters

Much of a team’s routine work depends on standards and preferences that live in people’s heads or are scattered across documents. Custom agents capture that approach once and apply it each time someone on your team runs the agent.

Because instructions describe outcomes rather than individual steps, agents are faster to build and easier to update when your standards, templates, or jurisdictions change. People running an agent also spend less time prompting it along, and scheduled agents can handle recurring work on their own cadence.

How to Use

You can learn more about creating, using, and managing agents in our Help Center articles below:

• To learn how to create, test, and refine agents, see Build Custom Agents.
• To learn how to find, run, and schedule agents, see Run and Schedule Custom Agents.
• Admins can assign agent permissions and manage workspace agents by following the instructions in Managing Agents.

FAQs

Q: How are custom agents different from workflows?

A workflow runs a set sequence of steps exactly as it was built. A custom agent works from text instructions and determines the steps for each run based on the outcome you describe.

Q: What happens to my existing workflows?

Existing workflows continue to run as built, and Workflow Builder remains available.

Q: When should I build a custom agent instead of asking Harvey in a thread?

A thread is suited to one-time work. Build a custom agent when the work recurs and should start from the same instructions, sources, vaults, and templates every time.

Q: Do shared agents need admin approval?

By default, sharing with specific individuals doesn't require approval, while sharing with your entire workspace does. Admins can change these defaults in your workspace.