Laufwerk
Available to try · 0.0.1-alpha.6

Expert work.
Now repeatable.

Turn the way you work into AI workflows.
Define the process. Let agents do the work.
Keep the checks and decisions in your hands.

Build your first workflow See it in practice

A request becomes
a checked result.

A useful workflow does more than ask an agent to try. It gives the work a process and a way to know when it is done.

First workflowA small code repair
The request

“The order total ignores quantities.
Fix the calculation. Keep the tests unchanged.”

AgentEdits isolated files
CheckRuns the tests
ResultReturns the change
total.mjs
− sum + item.price
+ sum + item.price * item.quantity
Tests fail? The workflow stops before write-back.Run this example

A disposable teaching fixture. You define the tests; their coverage determines what they prove.

Your expertise defines the process.
Laufwerk runs it.

Which sources matter? What makes a result acceptable? When should a person decide? Put those rules into a TypeScript workflow.

Laufwerk runs agent sessions with isolated working files, records execution state, and lets you inspect runs and answer human requests in a local Studio.

Understand the moving parts

Start with a job you know.

Code is one example. The same building blocks work wherever judgment, evidence and repeatable steps meet.

Keep documentation accurateRead the source. Repair the guide. Check the result.Investigate an incidentTurn logs into observations, likely causes and next checks.Make room for human judgmentPause for a decision. Continue with the answer recorded.

Know what you’re running.

Try the published package and reproduce the checks. Laufwerk is an early local runtime, with explicit limits and examples you can reproduce.

View the package ↗See what’s verified →

Built by Ralf Boltshauser, a member of the Agentic Systems Lab at ETH Zurich ↗.

Make one useful
thing repeatable.

Start with a small checked change.
Then make the workflow your own.

Build your first workflow Check the setup requirements first →