Your agents, in your pocket.
You already run Claude Code or Codex on a good machine. Patchkin lets you reach that machine from anywhere, hear about it when an agent needs you, and stop recreating your setup on every computer.
The moments Patchkin is for.
“It has been asking me a question for an hour.”
Agents stop at permission prompts. Patchkin marks a session as may be waiting for you when it goes quiet on a question, so you can open the app and answer from your phone instead of finding out at 6pm.
“I left the laptop at the office.”
Open the session’s live terminal in any browser. Read what it did, answer it, or send the next instruction.
“This task needs the big machine.”
Continue a session on another paired machine and keep going with the notes and recent output, instead of explaining everything again.
“I set up my rules on one machine only.”
Write a rule, command or skill once. It lands in the right place for each tool on each paired machine, and your own files are never overwritten.
What it looks like in practice.
Four moments, one place.
- Morning: launch the long job on your desktop, in the folder you choose, with a name you will recognise later
- Midday: the agent asks to run a command. You get the alert, open the terminal and approve it, or tell it no
- Afternoon: try Codex on the same repository from the same place. Your standards follow it
- Evening: see what ran, what changed and what it cost, without digging through terminal history
What your team can and cannot see.
Never collected
- The contents of your files
- Your prompts or anything you type
- The output of your tests and linters
- Secret values (a finding says what kind and where, not the value)
Recorded only when your organization turns governance on
- Which agent ran in which repository, and when
- Branch and commit names, changed file names and line counts
- Whether a rule held, for example a secret appeared in a diff
- Cost and token counts
You always see your own activity. Your organization must show you what is recorded, and per-person views are off unless they choose otherwise.
Honest details.
It is a small program
One daemon per machine. It connects out to Patchkin, so there is no port to open and no VPN. It runs on macOS, Linux and Windows.
You can still use the terminal
Patchkin does not replace your shell. Use agents the way you do now, or from the browser when you are away.
Some chats are view-only
You can browse earlier conversations from several tools and continue the ones the tool supports. App chats from the Claude and ChatGPT apps are not listed.
Developer questions.
Will it slow my agent down?
No. The agent runs locally as it does today. Patchkin streams the terminal to the browser and records small events such as a commit finishing.
Does it read my code or my prompts?
No. It records names, counts and results. See the privacy section above for the full list.
Can my manager see what I typed?
No. What you type is never recorded. Per-person activity views are optional and only show the recorded names, counts and results.
What if I use several tools?
That is the point. Launch any of them from the same place. Standards are written once for nine tools, and AI attribution recognises the marks each one leaves in commits.
Can I use it on a server or a cloud box?
Yes, on any machine you can install the program on. Cloud sandboxes run by other vendors are a different case, so ask us about yours.
Try it on the machine you use most.
Pair one computer and run your first agent from the browser.