How to Run a Chat Agent in Eve

Last updated: September 30, 2026

Overview

A chat agent runs from inside a matter and returns its output as a conversational response. Running one takes three steps: pick the agent, pick the documents it should read, click Run.

Context selection is the step that decides output quality. Everything else is mechanical.

Before You Start

  • You are working inside an active matter.

  • The chat agent exists in your agents library, either your own or one shared with you.

  • The documents the agent needs are uploaded to the matter.

Steps

  1. Upload the documents first. An agent with nothing useful to read still produces output, it is just generic. Confirm the files are in the matter's document panel before going further.

  2. Open the agents section in the matter. You see your own agents plus anything shared across the firm.

  3. Find the agent. Chat agents carry a chat bubble icon, which tells you the output comes back as a conversational response rather than a document. Search by name, or select See All Agents.

  4. Select the matter context documents. Choose the specific files Eve should analyze. Fewer, targeted documents produce faster responses and keep the analysis focused; irrelevant files introduce facts that muddy it. More documents help only when the task genuinely requires breadth. Default to relevance over volume: if the agent summarizes medical records, select the medical records rather than the whole file.

  5. Add Central Library context if the agent needs it. Select the relevant collection when the agent should reference firm-wide material such as style guides or objection banks. Skip it otherwise.

  6. Add a research collection if the agent needs it. Worth adding when the analysis turns on legal standards rather than case facts. Every piece of context added is more for Eve to weigh, so skip it when it is not needed.

  7. Click Run. Eve works through every prompt in the agent in sequence, each building on the last, with no further input from you.

The agent runs in the background. Move to another matter or start something else; the full output is waiting in the conversation when you come back.

Review the Output

Check the quotes and citations and confirm the key facts trace back to your documents, the same standard you would apply to any Eve response. This matters more with agents than with one-off prompts, because a shared agent runs the same way for everyone at the firm and an unverified habit scales with it. See 📄 How to Verify Eve's Responses.

Keep Working in the Conversation

When the agent finishes you are in a normal chat. Ask follow-ups, narrow the scope, change the format, request more support. See 📄 How to Refine Eve's Responses.

Re-running the agent with a different context selection is often the fastest fix when the output missed something.

If the Output Is Not What You Expected

Check in this order:

  1. Were the right documents selected?

  2. Were they actually uploaded to the matter?

  3. Only then, do the agent's instructions need work?

Context is the cause far more often than the agent is. If you own the agent and the instructions do need adjusting, see 📄 How to Edit Agents in Eve.