Self-hosted
Run it yourself: paste one prompt into Claude Code. The Druks chat walks you through the rest.
- Where Druks runs
- On your laptop or your own server
With your AI: paste this into Claude Code or Codex
Or in a terminal
Druks runs recurring engineering work with the AI tools you already use, then brings each proposed change to you for approval.
No install · Your own Claude or OpenAI account · Approve in Slack
Email me a link for my computeror Install Druks on your AIThis closes 7 vulnerabilities, 2 of them in code you call. Tests pass. No cost while I wait.
Tell me what to change. I will try again and ask you again. The plan now:
- "jsonwebtoken": "8.5.1", + "jsonwebtoken": "9.0.2", - "axios": "0.21.1", + "axios": "1.7.4", - "semver": "5.7.1", + "semver": "7.6.3",
permissions: block.A changed tag can run someone else's code in your build. I can pin each action to a commit and limit the token. Want a second pull request?

Our adoption of agentic operations stalled because we lacked an orchestrator that could run unattended and be trusted: deployed internally, fully auditable. Druks made that real for engineering, and it’s now reshaping how our Customer Success, Sales and Marketing teams operate.
Druks starts from a schedule, an alert or a button. It stops at the moment that needs a person, and asks you where you already are.
⏸️ acme/api: 3 package upgrades ready. Closes 7 vulnerabilities (2 in code you call). Tests pass.
Pull request #142 opened in acme/api. 3 files changed. Cost $0.44.
What Druks does with your work, and what it does not do.
The agent's workspace never holds your AI or GitHub token. It sees a placeholder. Druks gives the real token only for the call, and it keeps your secrets encrypted.
After a restart or a deploy, a run continues from its last saved step. Finished work does not run again.
When a run asks you, Druks closes its workspace and waits up to 14 days for your answer. No agent runs, so no AI cost.
Each question accepts one answer. A second click or a retry does not approve twice.
Each proposal, each answer and its cost stay in the run history.
Self-hosted, your code and your data stay on your machine or your server. You can read all of the source code. Report a security problem.
Each process is an app with its own schedule or trigger. Example schedule and run data.
| Process | Starts | Last run | Result | Needs |
|---|---|---|---|---|
| Security sweepFixes vulnerabilities and hardens your build | Every Monday 07:00 | Today, 14 min | 2 pull requests ready | Your yes |
| Ticket to pull requestWrites, checks and opens the change | Ticket labeled "druks" | Yesterday, 41 min | Pull request #188 | Approved by Anna |
| Your own processA Druks app is a Python package | You choose | — | — | Read the app guide |
Run it yourself: paste one prompt into Claude Code. The Druks chat walks you through the rest.
With your AI: paste this into Claude Code or Codex
Or in a terminal
We run it for you: sign up, connect your repositories and Slack. No install.
The rest is the same: your own Claude or OpenAI account, and you approve in Slack, the dashboard, the chat or Claude Code.
Druks is a self-hosted or cloud platform where AI agents such as Claude Code and Codex do a company's recurring engineering work, and a person approves each change.
Your own Claude or OpenAI account. Druks runs the official Claude Code or Codex tools with it. You see the cost after every step.
No. A run stops before anything changes in your repositories and waits for your decision, up to 14 days. It costs nothing while it waits.
Druks is source available under the Functional Source License (FSL). You can self-host it, change it and build apps on it. Each release becomes Apache 2.0 two years after it ships.
Druks comes with processes that already run, like the security sweep. The work starts by itself from a schedule or an alert, uses your own AI tools, and reaches you as one decision.