Run agents anywhere you have a machine.
Pair your laptop, desktop or a server, then start and steer agents from a browser.
- Launch from the web. Pick a machine, a folder and Claude Code, Codex, Aider or a shell, give the session a name, and start it
- Live terminal. A real terminal in the browser. Take the keyboard, release it, or hand it to the agent
- Continue elsewhere. Start the same work on another paired machine with the notes and recent output, and Claude Code’s conversation comes along. It is a new process: the folder and tools must already be on the other machine
- Past conversations. Browse what Claude Code, Codex, OpenCode, Aider and Cursor have done on a machine, and resume the ones the tool can pick back up. Cursor conversations are view-only
- Any other terminal tool. Run Cursor CLI, OpenCode, Gemini CLI or anything else in a shell session
Watch what is happening.
See every machine and session, and get told when something needs a person.
- Machines at a glance. Online or offline, operating system, version, how busy, and which agent tools are installed
- Waiting alerts. A session that goes quiet on a question is marked may be waiting for you, and the app shows the question. Optional browser notifications say that a session is waiting; the question text is not sent in them
- Activity and analytics. A timeline of what happened and trends over time, per team
- One-click updates. See when a machine is behind and update it from the web
Standards, written once.
Rules, commands and skills for your team, delivered to each tool where it expects them.
- Nine tools. Claude Code, Codex, Cursor, OpenCode, Aider, Gemini CLI, Copilot, Kiro and Amp
- Every paired machine. A machine picks up a change as soon as it is online. Tools that are not installed are skipped
- Your files stay yours. Shared files only get a clearly marked block. Withdraw a standard and the block is removed
- By team. Choose which teams a standard applies to, and turn one off without deleting it
What works with which tool.
We have tested Claude Code end to end. The others follow each tool’s documented behavior and have not all been tested with the live tool yet.
| Tool | Launch from the web | Past conversations | Standards | Marks AI work in commits |
|---|---|---|---|---|
| Claude Code | Yes | Browse and resume; moves with Continue elsewhere | Rules, commands, skills | Yes (co-author trailer) |
| Codex | Yes | Browse and resume | Rules, commands | None known: counted only when run through Patchkin |
| Aider | Yes | Browse | Rules, through a conventions file | Yes (“(aider)” in the author name) |
| OpenCode | In a shell session | Browse and resume | Rules, commands | None known |
| Cursor | In a shell session (Cursor CLI) | Browse only | Rules and commands, written as files in your repositories | Yes (cloud agents add a co-author trailer) |
| Gemini CLI | In a shell session | No | Rules, commands | None known |
| Copilot CLI, Kiro, Amp | In a shell session | No | Rules | Copilot: yes. Others: none known |
A shared context bank.
Keep the notes, commands and background your team reuses, and put them into a running session.
- Reusable context. Store what you keep explaining to agents
- Drop it in. Inject it into the session you are driving, so the agent starts informed
Notes and commands
The things every new session needs to know about your codebase.
Into any session
Inject into the session you are driving, with one click.
Governance.
A record of AI work and rules that are checked. Details on the governance page.
- Sessions and commits. Who ran which agent in which repository, with branches, commit names and changed file names and counts
- Findings. Secrets, source without tests, force pushes, protected branches, CI file edits, skipped permission prompts, skipped hooks, large deletions, hidden text in instruction files
- AI attribution. AI-assisted commits and pull requests, recognised from the marks tools leave
- Checks at session end. Optional test or lint runs, recording pass or fail only
- MCP and instruction inventory. Which servers and instruction files each machine uses, with an approved list
- Enforcement. Git hooks on paired machines, and a required check in GitHub, GitLab, Bitbucket, Azure DevOps or Jenkins
Usage and cost.
Know what agents cost, and what a merged pull request costs.
- Counts only. Cost and tokens, never prompts, responses or code
- By tool, person and repository. Over a period you choose
- Cost per merged pull request. Using merges from your connected git host
- Claude Code today. Sessions started in Patchkin report on their own. Other tools are not supported yet
What an organization needs to roll this out.
Single sign-on
OpenID Connect and SAML.
User provisioning
SCIM, so people are added and removed from your identity provider.
Two-factor and an activity log
Two-factor sign-in, and a log of who did what in the product.
Roles and teams
Owners, admins and developers, across teams within an organization.
End-to-end encryption
Optional, per machine. Patchkin’s servers cannot read an encrypted terminal.
Platforms
macOS, Linux and Windows. The program connects out only.
See it on your own machines.
Pair one machine, start an agent from your browser and watch it work. Tell us about your team and we will set you up.