Privacy · last updated 2026-11-01

Privacy

Jimmy runs on your own PC. It has no account, no server of ours and no telemetry. Its author never receives your captures, your questions or your keys.

From the trailer

What Jimmy stores

All of it is in one data folder on your PC only: the folder you picked on first start, or data inside the Jimmy folder for a source install.

  • A capture database: window app and title, window text, transcribed speech near your mic, and the number of faces in each frame.
  • Blurred screenshots, about 80–88 KB each.
  • Jimmy's own database: things you asked it to remember, reminders, goals, conversation turns and a decision log of each request.
  • Your wiki: short pages about your projects, interests and habits, built from the above, which you can confirm or delete.
  • A log of the console, where captured text appears only as its length, never as text.
  • Keys you typed into the installed app's setup, encrypted with Windows DPAPI. (A source install reads keys from .env or your environment.)
  • Only if you say "remember my face": your face template, encrypted with Windows DPAPI so only your Windows account on this PC can read it.

What Jimmy does not store

Unblurred screenshots, raw audio, webcam frames, other people's face data, anything from an excluded window (password managers, banking and payment sites, private browser windows), and system audio such as the other side of a call. The details are on the security page.

Who processes data

  • The model provider you choose (NVIDIA, Google or OpenAI), only if you add a key, receives your question and the evidence for its answer, a screenshot of the window for a question about your screen, and the other items listed under what leaves your PC. Each handles that data under its own privacy policy. On Google's free tier, Google may let reviewers read it.
  • GitHub, from version 1.0.1: every six hours the installed app asks for the latest release and, if it's newer, downloads the setup. That is a normal web request with nothing about you in it. Settings turns this off.
  • Another assistant you connect through the MCP server receives what its recall tool returns.
  • During setup and first start, Jimmy downloads its own Python (through the uv tool), its packages from the Python package index, the Whisper speech model from Hugging Face and, with the Full profile, Ollama and its models. A source install also uses npm. Each of these sees a normal download request.

Ollama, the card model, meaning search, speech recognition and the face detector all run on your PC.

Your choices

  • Pause with Ctrl + Alt + J; nothing is stored while paused. The curtain stops capture when you leave.
  • Forget any span: "Jimmy, delete everything from September" shows what goes and asks first.
  • Exclude more windows in data\exclusions.txt.
  • Keep everything local by running without a model key.
  • Limit it in the installed app's Settings: how many days to keep (30 by default) and a size cap (20 GB by default).
  • Delete it all by deleting the data folder; uninstalling asks whether to. "Forget my face" removes the face template.

Because nothing reaches us, there is nothing for us to delete on request.

This website

This site sets no cookies and runs no analytics or trackers. It loads its fonts from Google Fonts, so your browser contacts Google for them, and it is hosted on Render, whose servers handle each request as any web host does.