Micro agents that maintain and build your system while you sleep.
Each agent does one job: fix the pipeline that broke at 3am, update a dependency, add the test nobody wrote. It works through the night and leaves you a merge request to review in the morning.
Runs on the Selan account you already have. Already on Selan? Open reactor.selan.ai.
Small agents, one job each.
Give it one job
A short prompt, a model and an effort, and only the connectors that job needs. A small agent is easy to read and cheap to run.
Decide what wakes it
A time of night, an event from GitLab or GitHub, a Slack reply, or another agent that is allowed to start it.
It works on its own
Up to four hours on a fresh VM with Docker, Chromium and PostgreSQL 18, or on a machine of yours. It remembers what it learnt last time.
You review in the morning
A merge request, a Slack post or a one-line status on its card, with the full transcript behind it. You decide what merges.
Every action is an MCP tool, for you and for your agents.
Run the fleet from Claude Code
Every Reactor action is a Selan MCP tool: make an agent, give it a trigger, read its runs, sort the canvas, message a run that is still going.
> make an agent that adds a missing test every night ● selan_remote_agent_create ● selan_remote_agent_trigger_create
Each agent gets the connectors you give it
GitLab, Slack, Jira, Postgres, Google Cloud, even a tab in your own Chrome. Connect them once for the company, then hand each agent only what its job needs.
Agents call each other
Triage hands a broken docs page to Docs review and waits for the answer. An agent can start another only if that agent lists it, and Reactor refuses a start that would loop.
Run them anywhere.
A fresh VM per run
defaultMade for the run and thrown away after. Nothing a run does outlives it.
Your own Google Cloud
your quota, your billStill a fresh VM each run, made in your project instead of ours.
Your own runner
starts in secondsA Linux server or an Apple Silicon Mac you keep on, installed from one line. A Mac can build with Xcode.
Wherever it runs, an agent spends through your Selan gateway, so its model calls show in your spend next to everyone else's. Selan keeps each run's transcript for 90 days, and each agent's memory until the agent changes it, in Selan's Google Cloud in the EU.
Micro agents that maintain and build your system while you sleep, and a morning where you only review what they did.
Triggers and memory
Agents that start themselves and remember the last run.
Workspaces and groups
A fleet you can read, sorted by team and by job.
Supervisors
An agent that tunes the others in its workspace.
Whole codebases
Fleets that keep a system current and working, with people steering.
Put a night shift on your codebase.
Reactor is in early access. We switch it on one company at a time, on the Selan sign-in you already use.