Agents in Eve
Last updated: October 23, 2026
An agent is a saved set of instructions that tells Eve what to do and, optionally, when to do it. Agents live in the Agents section and run against the documents in a matter.
Every workspace arrives with example agents already built. Firms can add their own, keep them private, or share them so a workflow one person worked out becomes the way the whole firm does that task.
Agents replace what used to be Playbooks and Blueprints. If you used Eve before the change to agents, your playbooks and blueprints were automatically updated to an agent, no changes needed.
How It Works
An agent takes an instruction or task on for itself, helping to handle firm needs and start tasks. Eve has four kinds of agents:
Chat agents return a conversational response: analysis, summaries, evaluations, alerts, and recommendations. See 📄 Chat Agents
Drafting agents return a formatted document, saved to the matter or exported. They learn your firm’s structure and language from 2 to 5 example documents, and your instructions are saved so you can reuse them. Demand letters, complaints, and mediation statements are common uses. See 📄 Drafting Agents
Discovery response agents are drafting agents tuned for written discovery: interrogatories, requests for production, and requests for admission. See 📄 Discovery Response Agents​
Communication agents place outbound voice calls to clients and insurers. The agent builds a call plan from matter context for you to review and approve, places the call, navigates menus or voicemail, and returns a transcript and summary. Treatment check-ins, records follow-ups, and status updates are common uses. Nothing goes out until you approve the plan. See 📄 Communications Agents Overview
What starts an agent
Manually. You open a matter and run the agent. This is always available.
On a trigger. The agent runs on a schedule or in response to matter activity, such as a new document upload, with no one present. An agent set up this way is an auto agent. See the Auto Agents and Triggers article.
The two aren’t exclusive. Setting a trigger doesn’t remove the option to run the same agent by hand.
Note: Medical Overview, Eve’s built-in medical records summary, can’t currently be set up as an auto agent. You can still run it manually any time.
Context Is What Decides Quality
Every agent reads the matter context selected for the run. Anything left unchecked may as well not exist for that task, and anything never uploaded can’t be read or cited at all. You can also add firm reference material, like a style guide or standing instructions, from the Central Library alongside your matter context.
An agent’s output is a first draft. It comes with citations back to the source documents, so check the claims that matter against them before anything leaves the firm.
Key Terms
Agent: A saved set of instructions defining what Eve produces and what starts it.
Chat agent: An agent that returns a conversational response rather than a document.
Drafting agent: An agent that returns a formatted document, built from example documents supplied by the firm.
Discovery response agent: A drafting agent tuned for structured responses to written discovery.
Communication agent: An agent that places outbound calls to clients and insurers, with an approval step before any call goes out.
Manual agent: Any agent run on demand by a person. This is always available.
Auto agent: Any agent set to run on a schedule or a trigger rather than on demand.
Trigger: The condition that causes an auto agent to run.
Matter context: The documents Eve reads during a run. Unchecked files are excluded entirely.
Central Library: Firm-wide reference material an agent can read alongside matter context.