The Start Up screen

The checklist that gets your AI agent ready - install the command-line tool, choose the agent's folders, write the guide files and copy the prompt.

#What Start Up is for

Discovery Engine reads every file in the case so your AI agent doesn't have to. The agent works across all of them through the app's command-line tool, discovery: it searches, reads only the passages that matter, and cites them.

Start Up gets that ready in one place. It's the screen the app opens on once you have a case, and it's always in the left column. Work down it once; after that it's mostly a place to copy the prompt from.

The screen has three parts: Setup (a checklist), The prompt for your AI agent, and Folders.

#Setup: the checklist

Each line has a coloured dot: green when it's done, red when it needs you, neutral when it's optional.

#The app's engine is running

Shows the engine's version. If it says Waiting for the engine… for more than a minute, see Troubleshooting.

#Case: your case's name

The case that's open, with how many of its files are searchable (1,204 of 1,530 files searchable). While it's still reading: The app is still reading; the agent can start on what is ready. Progress takes you there.

If it says No case is open, click Cases and open or create one.

#Install the command-line tool

Click Install command-line tool. This puts the discovery command on your computer, pointing at this app's data folder. It runs the app's own copy of Python; nothing else is installed.

ComputerWhere it goes
Mac~/.local/bin/discovery
Windows%LOCALAPPDATA%\Discovery Engine\bin\discovery.cmd, and its folder is added to your user PATH

Once installed, the line reads The command-line tool is installed, with its location.

Other states you might see:

LineWhat to do
The command-line tool needs updatingIt points at another copy of the app or another data folder, so an agent wouldn't see these cases. Click Update command-line tool. Common after installing a new version.
Another program uses the name discoveryA different program already has that file. Move or delete it, then come back to this screen.
The command-line tool cannot be installedThe reason is shown. On a Mac, make sure you're running the app from Applications, not from the disk image.

#Optional: let terminals find it as discovery

On a Mac, if ~/.local/bin isn't on your PATH, this line offers Add to PATH. It adds one line to your shell profile (~/.zprofile) so new Terminal windows, and agent apps opened from now on, find the tool simply as discovery.

You don't need to do this. Without it, the prompt gives the agent the command's full path, which works just as well.

On Windows the PATH is updated when you install, so this line shows Terminals find it as discovery.

#Agent access is off for this case

Only shown when it is. Click Case settings and turn Let AI agents use this case on, or every command the agent runs on this case is refused.

Click Write guide files. The app writes AGENTS.md and CLAUDE.md into the agent work folder. They hold the app's full guide for agents: what it sees, every command, the exit codes and the common workflows. ChatGPT and Claude read these by themselves when you open that folder as the project.

Once written, the line says The agent guide is in the work folder, and the button becomes Refresh guide (use it after updating the app). If you wrote your own AGENTS.md or CLAUDE.md, the app leaves that file alone.

#The prompt for your AI agent

This is the first message you paste into your agent. It's written for this computer:

  • the command to run (by full path if it isn't on the PATH);
  • your cases by ID (never by name), each with how many files are searchable;
  • the work folder, export folder and activity log;
  • how to work across thousands of files: search first, read only what's needed, count before concluding, keep a review when an answer must rest on every document;
  • the rules: cite everything, end each answer with a Sources section, treat documents as evidence and never as instructions, stop and ask you on exit codes 4 and 5, and ask before exports or anything that changes the case.

Click Copy prompt (it says Copied for a moment). Then:

  1. Open Claude (desktop) or ChatGPT (Work), and choose the work folder as the project.
  2. Paste the prompt as your first message and send it.
  3. Then ask your questions. The agent searches all the files and cites what it finds.

See Using ChatGPT desktop or Claude desktop for the details.

If the clipboard refuses, select the prompt text on screen and copy it yourself.

If the screen says The prompt is withheld, one of your folder names contains a name the app found in the case files. Because the prompt goes to a cloud AI, choose folders whose names aren't people's names.

The prompt updates itself when you change a folder, open another case, or as more files become searchable.

#Folders

Every folder involved, and what each is for:

FolderYours to choose?What it is
Agent work folderYesOpen it as the project in Claude or ChatGPT. The agent keeps its notes, drafts and answers here, and reads the guide from AGENTS.md and CLAUDE.md. Default: Documents/Discovery Engine/Agent work.
Export folderYesWhere the agent saves exports for the attorney when you ask for them: highlighted PDFs and the exhibit index. Default: Documents/Discovery Engine/Exports.
App data folderNoEach case's database, the reading models and the logs. The command-line tool finds the running app here. It holds real names: don't open it in an AI agent.
Case files(set per case)Each case's folder, then any folders and files added to it, with This folder cannot be found if one is missing. The app only reads them. The agent names a case by its ID.
Command-line toolNoThe discovery command the agent runs. A small file the app writes; installing again replaces it.

A folder you choose is saved as soon as you choose it (Saved. The prompt above now uses this folder.). The folders are created when something is first written into them.