LoopLoop

Loop Documentation

Loop is the autonomous improvement engine that closes the feedback loop for AI-powered development. It turns signals into prioritized, actionable work — and tells AI agents exactly what to do next.

Loop Documentation

Loop is the autonomous improvement engine — an open-source data layer and prompt engine that collects signals, organizes work into issues, and tells AI agents exactly what to do next.

It closes the feedback loop for AI-powered development by turning raw data (errors, metrics, user feedback) into prioritized, actionable work items.

Signals  ->  Triage  ->  Hypothesize  ->  Plan  ->  Execute  ->  Monitor
  ^                                                                 |
  +-----------------------------------------------------------------+

Loop has no AI

Loop is a fully deterministic data system with a prompt layer. There are zero LLM calls inside Loop — no embeddings, no inference, no token costs, no model dependencies. AI agents pull work from Loop over HTTP. When a better model ships, every Loop installation gets better overnight without changing a single line of code.

How It Works

  1. Signals arrive — webhooks from PostHog, Sentry, GitHub, or any HTTP source create issues for triage
  2. Agents triage — an AI agent picks up the signal, decides if it is real or noise
  3. Hypotheses form — the agent creates a hypothesis with confidence level and validation criteria
  4. Plans break down — the hypothesis becomes concrete, scoped tasks
  5. Agents execute — tasks are dispatched one at a time with rich context and instructions
  6. Outcomes feed back — monitoring tasks watch for results, creating new signals that restart the loop

The issue queue is the orchestration layer. One unified system, one priority system, full auditability, and human override at any point.

Explore the Docs

Key Principles

  • Agent-agnostic — any AI agent that can make HTTP calls works with Loop. Swap agents or models without touching Loop.
  • Pull architecture — agents poll for work on a schedule. Loop never needs to know how to connect to your agent platform.
  • Self-improving instructions — agents rate the prompts they receive. When quality drops, Loop creates an issue to improve them.
  • Everything is an issue — signals, hypotheses, plans, tasks, and monitors all live in the same priority queue.
  • Open source — MIT licensed, fully on-premises capable, no vendor lock-in.