Open-source agent runtime
Run deep agents on your PCs
Run a blueprint directly, see every step, and keep the data fully yours.
MIT licensedmacOS, Linux & WSL2Docker required
At a glance
- Run a blueprint
- Start directly
- State, retries, recovery
- Durable execution
- Local, private, edge
- Deployment
- MIT
- License
How it works
A small runtime for work that keeps going.
The agent code remains yours. MirrorNeuron handles the lifecycle around it.
- 01
Start with a blueprint
Run a blueprint directly, then adapt its code and tools to your work.
- 02
Let it keep going
MirrorNeuron persists state, retries failed steps, and resumes from known checkpoints.
- 03
Stay in control
Inspect what happened, pause for approval, and choose exactly where the runtime operates.
Choose the smallest tool that fits
Durable execution without the orchestration project.
One-off script
Fast to start
You own state and recovery
Airflow or Temporal
General orchestration
A larger platform and vocabulary
MirrorNeuron
Durable agent work
Normal code and a small runtime
Use cases
Work that outlives a chat.
Background agents
Research, monitor, call tools, wait, and resume without keeping one process alive.
Learn morePrivate analysis
Keep workflow state close to financial, scientific, or regulated data.
Learn morePhysical and edge AI
Run near sensors, video, machines, and local models when latency and ownership matter.
Learn moreStart locally
Two commands to see it work.
Install the runtime, launch a complete workflow, then inspect and change the code at your own pace.
>_1. Install MirrorNeuron
$ curl -fsSL https://mirrorneuron.io/install.sh | bash>_2. Run a blueprint
$ mn blueprint run vc_assistant