Skip to content

Desktop App

A window for running reviews without the terminal. It runs the same pipeline as the CLI, on your machine.

Install

CI builds the installers (.github/workflows/desktop.yml) and attaches them to the GitHub Release of a version tag.

System File
macOS (Apple Silicon) .dmg. Drag the app to Applications
Windows (x64) -setup.exe. Installs for your user, no admin rights

On macOS you can also use Homebrew:

brew install --cask zoharbabin/due-diligence-agents/due-diligence-agents

Homebrew 7 asks you to trust a third-party tap first. Run the brew trust command it prints.

Release builds are signed. The macOS app is also notarized. The app checks for a newer signed version when it opens and offers to install it.

First run

  1. Open the app. It starts its own local service.
  2. Download the Claude engine. One click. The app downloads Anthropic's claude command line tool and checks it against a fixed digest before use. It is stored under ~/.dd-projects/.cli/ (or $DD_BASE_DIR/.cli/). A claude already on your PATH is used instead.
  3. Paste your Anthropic API key. The app tests it and stores it in your system keychain. For Bedrock, Vertex or a gateway, set the provider variables before you start the app. See Model Providers. To test, replace or remove the key later, open Settings.
  4. New project. Pick the folder with your contracts and run the review.
  5. Ask questions. Open Ask after a review. Answers come from the findings of the latest review and its documents. Each question calls the model and costs money. The page shows the session cost against its limit. New conversation ends the session. A newer review starts a new session.
  6. Optional: tune an agent. In a project, open Agents. Pick one, start from a profile and save. The change applies to that project only and is checked before it is saved. Use Preview prompt to see exactly what the agent receives. The safety rules cannot be changed.
  7. Compare reviews. After a second review, open Compare to see new, resolved and escalated findings. It uses the same engine as dd-agents diff.
  8. See what happened. Open History for a list of reviews and questions, newest first, with findings, cost and entities. You can filter by type.
  9. Screen a name. Open Screening and press Update lists to download the official US, UK and EU sanctions lists (needs the internet, stored on your computer). Then check one company or person. Results are possible matches for a person to review. No match is not a clearance. It uses the same engine as dd-agents screening. See screening.
  10. Let the model draft the setup. In Settings, save the buyer and target names, then press Suggest a setup. The model reads your folder and file names and drafts the configuration. This costs money. The draft opens in the editor and is not saved until you press Save. It uses the same steps as dd-agents auto-config.

Where things are

What Where
Projects, findings, reports ~/.dd-projects/ (or $DD_BASE_DIR). Same as the CLI
API key OS keychain, service com.zoharbabin.due-diligence-agents
Claude engine ~/.dd-projects/.cli/<version>/
Service log (errors from the last launch, replaced at each launch) macOS: ~/Library/Logs/com.zoharbabin.due-diligence-agents/service.log. Windows: %LOCALAPPDATA%\com.zoharbabin.due-diligence-agents\logs\service.log

Security

  • The service listens on 127.0.0.1 only. Every request needs a token that is made at each launch and never written to disk.
  • Opening the app a second time shows the open window. It does not start a second service.
  • The service stops when the app closes, with all of its child processes.
  • Your documents stay on your machine. Only the text sent to the model provider you chose leaves it.

Develop and build

See desktop/README.md. The API is described in dd-agents serve.