The AI that runs your
portfolio._
A self-hosted daemon that schedules Claude agents, watches deploy health, handles PR factory duty, and keeps your 15 side projects from rotting while you build the next one.
/kruz
See the daemon running in production on one founder's actual portfolio. Job feed, watchdog status, deploy health, recent sprint report.
What the daemon does while you sleep.
Scheduled Agents
Recipes fire on cron. Claude agents run in worktrees, open PRs, tear themselves down. You review the diffs, not the process.
Watchdog
Every status section has a max-age. When a job hasn't run in N hours, Operator pings Discord before you notice something stalled.
Discord Surface
!op status, !op morning, !op review prs, !op deploy check. Slash commands too. Your whole portfolio ops from one channel.
Claude Code Hooks
Pre/post tool-use guardrails, risk tiering, auto-merge gating. Protected paths never merge without explicit approval.
PR Factory
Worktree per task. Agent works in isolation, runs your test suite, opens a PR. Green CI auto-merges if you opted in.
Observability
Local /ops dashboard, Prometheus /metrics endpoint, job ledger in sqlite. Everything the daemon does is inspectable.
Swap 20 minutes of tab-hopping for one paragraph.
7 tabs, 20 minutes, still not sure what to do first.
- [tab]Vercel dashboard — which deploy failed last night?
- [tab]GitHub notifications — what PRs are open?
- [tab]Supabase logs — anything broken at 3am?
- [tab]Gmail — who needs a reply?
- [tab]Linear / Notion — what was I working on?
- [tab]Stripe — any new signups or failed charges?
- [tab]Discord / Slack — did anyone page me?
You still have to assemble the picture yourself. Every morning.
One Discord message at 7am. Read it in two minutes.
Two of seven projects moved overnight: prospector-pro shipped the enrichment retry, deal-brain has a PR open that's been sitting 4 days. Vercel is all green. Claude spend held at $0.34. Nothing paged — your machine is quiet. One move for today: merge the deal-brain PR so the enrichment work can rebase on top of it.
Same daemon, your rules. The brief is one Claude call on real portfolio data — git log, deploy events, spend ledger.
Your machine. Your agents.
# install and initialize
$ pip install operator-core
$ operator init
# edit ~/.operator/config.toml, add your projects
$ operator doctor
config path: /home/you/.operator/config.toml
projects: 12
doctor: OK
# start the daemon
$ operator run
[operator] bound 127.0.0.1:8765
[operator] scheduler tick: 6 tasks registered
[operator] discord bot online_Pre-alpha waitlist.
Not on PyPI yet. We're inviting ~10 builders from the waitlist per week. Early users shape the recipe library.