Overview

loopkeep runs personal AI automation workflows on your own machine, supervised from one inbox. A resident daemon watches your triggers — cron, file changes, git events, finished runs, and the things that happen to you on GitHub and Slack — and starts workflows without you kicking them off. Policies decide which actions run unattended and which wait for your approval; everything that needs your judgement lands in the inbox. You drive it from whichever surface you reach for — the lk CLI, the terminal UI, or the desktop app — all equal, all talking to the same daemon.

How it fits together

ConceptWhat it is
WorkflowA markdown file that tells the agent what to do. Compatible with GitHub Agentic Workflows.
TriggerWhen a workflow fires: cron, file changes, git events, finished runs, GitHub and Slack events, your own webhooks, or by hand.
RunOne execution of a workflow, checkpointed step by step in its own git worktree.
PolicyDeterministic rules that route each proposed action to auto-approve, notify, ask-first, or deny.
InboxThe one place where approvals and notifications reach you.
ConsoleAn optional web control plane for pairing devices and connecting GitHub and Slack. Your code and secrets stay on your machine.

Start here

  1. Install — the standalone runtime installer or the desktop app; both deliver the daemon, lk, and the adapter.
  2. Quickstart — from an empty folder to a supervised run.
  3. The desktop app — a tour of the window, projects, and every setting.
  4. The terminal UI — the same dashboard, right in your terminal.
  5. Pair with the Console — optional; enables devices and remote triggers.

Guides

  • Workflows — authoring, structured output, budgets, chaining.
  • Triggers — every trigger type, filters, coalescing, chain limits.
  • Policies & trust — attention levels, layers, floors, policy explain.
  • Runs & the inbox — the run lifecycle, checkpoints, and intervening.
  • Take-over & hand-back — drive a run's session yourself, then return it.
  • Secrets — keychain storage, references, redaction and its limits.
  • GitHub Agentic Workflows — import, export, and the compatibility contract.
  • Agent skills — teach your coding agent to operate loopkeep with you.
  • The Console — devices, connectors, and what never leaves your machine.

Reference