Kombinat
Run AI agents on a schedule from Markdown files in your repo
FreeOpen SourceAICode6 impressionslaunches Nov 171 comment
Comments
>log in to commentHey everyone 👋 I'm Dan, the maker. My real story behind this tool: I wanted background agents working 24/7 at my company that handle on-call incidents, fix bugs, monitor support chats and do other useful work at night. I started on my laptop with the default Claude Code routines. My old Macbook Air did a great job, it worked well, and everyone wanted the same. But at that moment, it didn't scale to a team: a routine there is basically a text area. I wanted one central repo that everyone can contribute to, with the routines running on a Mac mini that we control remotely. I didn't find anything that fit, so I built Kombinat. What works today: cron schedules, webhooks, Claude Code, Codex or any other CLI agent, a web dashboard and a terminal one with full transcripts, and auto-pull, so merging a PR is how you ship a routine. What's missing: it is macOS only for now (it uses launchd), and runs happen only while the Mac is awake. I'd love to hear which routines you would run first, and what stops you from trying it. Demo with sample data: https://danilqa.github.io/kombinat/ Quickstart: https://github.com/Danilqa/kombinat#readme
Kombinat runs background AI agents on a schedule. A routine is a Markdown file in your repo with a schedule and a prompt. An always-on Mac runs it with Claude Code, Codex or any other CLI agent.
Who it is for: teams that want agents to handle on-call incidents, fix flaky tests or clean up dead code overnight, where anyone can add a routine with a pull request.
- Version-controlled. Merge a .md file and it goes live, no restart.
- Your own setup. Agents use your subscription login, skills and MCP servers.
- Dashboard. Every run with its full transcript, in the browser, the terminal or on your phone.
- Webhooks. A Linear issue or a GitHub event can start a routine.
How it differs: Claude Code's own Routines worked for me alone, but each one is basically a text area. Kombinat keeps routines in one Git repo that the whole team reviews and edits.
Open source (MIT). macOS only for now. Demo with sample data, nothing to install: danilqa.github.io/kombinat
More Open Source tools →74 free dev tools →
- Dead simple, lightweight engine for 24/7 background AI agents
Kombinat runs Claude Code, Codex or any other CLI agent 24/7 on your Mac, on a schedule or a webhook, from markdown files in your repo.
- for
- Teams and solo developers who want AI agents to do routine work in the background, on the Claude or Codex plan they already have.
- pricing
- open source
- license
- MIT
Key features
- Markdown routines — Define cron schedules or webhooks in markdown files with front-matter.
- Multiple agents — Run Claude, Codex or any shell command per routine.
- Version-controlled — Routines live in your repo; merge a .md file and it becomes active within minutes.
- No extra API costs — Agents run on your existing Claude/Codex subscription, not on separate API keys.
- Web Dashboard & TUI — View run history, calendar and transcripts in a web UI or terminal UI. Works amazing from your mobile
- Webhook support — Trigger routines via POST requests; request body is added to the prompt.
- Lightweight install — Just two dependencies, runs as a background daemon on macOS with Node.
- Remote access — Run the dashboard on your phone over Tailscale.
Use cases
- Automatically draft pull requests for issues labeled ‘agent’ overnight.
- Fix flaky end-to-end tests in CI each weekday.
- Clean up dead code and feature flags weekly.
- Respond to production alerts by proposing fixes via a webhook.
- And basically anything else you can write as a prompt, from release notes to weekly reports.
Kombinat vs alternatives
 Kombinat |  n8n |  Mastra |  Claude Code |  LangGraph | |
|---|---|---|---|---|---|
| Best for | Background AI agents on your subscription | Visual workflows across many apps | Building custom agents in TypeScript | Hands-on coding with an AI agent | Custom stateful agents in code |
| Pricing | Open source | Subscription | Free | Subscription | Free |
| DevHunt upvotes | 1 | 0 | 0 | 0 | 0 |
| Launched | Nov 2026 | — | — | — | — |
- Kombinat vs n8n: n8n has you wire nodes on a canvas. In Kombinat you write a prompt, and Claude Code or Codex does the job with your own skills and MCP servers, at no extra cost.
- Kombinat vs Mastra: Mastra means coding agents in TypeScript. Kombinat routines are plain markdown anyone can write, and agents use your own skills and MCP servers, on your subscription.
- Kombinat vs Claude Code: Claude Code's own scheduled tasks run in the cloud for one account. Kombinat runs it on your Mac with your full local setup, from routines your team shares, plus Codex.
- Kombinat vs LangGraph: LangGraph means building and hosting agent graphs in code. In Kombinat, a routine is one markdown file: merge it and it runs, with your own tools, on your subscription.
Kombinat FAQ
What do I need to run Kombinat?+
A Mac that stays on, Node 24, and the agents you use, such as Claude Code or Codex, signed in. Run pnpm kombinat install once, and Kombinat keeps running after reboots. If the Mac was asleep, each routine catches up on its last missed run, up to 7 days back.
How much does it cost?+
Nothing. Kombinat is free and open source under the MIT license. Claude Code and Codex run on the subscription you already pay for, not on API keys, so there are no per-token bills.
Which agents can I use?+
Claude Code, Codex or any other CLI agent that reads the prompt from stdin. They run with your usual setup, so your global skills, MCP servers and CLAUDE.md are picked up.
How do I add a routine?+
Add a markdown file to the routines folder. The frontmatter sets the schedule and the agent, and the body is the prompt. No restart needed. In a git repo, merge it to main, and the daemon pulls it within 5 minutes.
Can other apps start a routine?+
Yes. Add webhook: true, and Linear, GitHub or any other app can start the routine with a POST request. The agent gets the request body in its prompt. An optional gate command checks signatures and skips unwanted events.
How do I see what the agents did?+
Open the dashboard at localhost:7433, or run pnpm kombinat tui in the terminal. You see every run with its transcript, a week calendar and the daemon status, and can start any routine by hand. Over SSH or Tailscale, it works from your laptop or phone.
Edited by the makers, based on danilqa.github.io · Nov 10, 2026. Details may change; check the official site.
