How is Loopy different from a normal coding agent?

A normal coding agent often stops at failed tests or asks you to inspect the app manually. Loopy keeps the build loop running: code, tests, browser review, patch, repeat.

Your coding agent stops. Loopy keeps testing.

Q: Where do API keys go?

In ~/.loopy/config.json on your machine — never uploaded anywhere. Configure them via the local UI at loopy serve.

Q: Can I try Loopy without API keys?

Yes — loopy run --demo runs the full pipeline in demo mode with scaffold output.

Q: Can I steer the agent mid-run?

Yes. Manual takeover lets you queue steering messages during an active build; Loopy injects them into the next agent turn.

Q: How long can Loopy run unattended?

The default turn budget is 200; typical unattended runs last up to nine hours. Use the overnight queue and scheduler for portfolio builds.

Loopy is a local-first build loop for indie devs. It runs your coding agent, boots the app, runs tests, opens pages in a real browser, patches failures, and repeats until the product actually works in your repo.

Become a tester →

Loopy · loop local · your keys

Your idea

Waiting for your idea…

The build loop

Failure is not the end state. It is the next patch.

Loopy turns failed tests, broken routes, console errors, and missing setup into the next agent turn. It does not stop with advice; it keeps editing the repo and checking the app.

Backlog3

mediumLPY-114

Repair flaky checkout tests

Flaky-Test Killer

lowLPY-118

Bump deps + changelog

Dependency Bot

In progress2

highLPY-111

Add Stripe webhook retries

Codex · working

mediumLPY-109

Migrate auth to sessions

Claude Code

In review1

highLPY-106

Fix N+1 in dashboard query

3 files · +84 −19

Done7

lowLPY-103

Ship release notes v0.4

✓PR merged

mediumLPY-101

Document onboarding flow

✓PR merged

Tests drive the loopPytest, npm tests, and failing commands become concrete repair instructions for the next coding turn.

Browser checks are first-classLoopy boots the app, opens routes, captures screenshots, watches console errors, and loops on real UI failures.

Repo-native outputIt works in your folder, reuses your GitHub repo, commits/pushes changes, and leaves a traceable run receipt.

Failure recoveryIf one pass breaks, Loopy feeds the failure back into the agent instead of handing you a TODO list.

Local by defaultYour code stays on your machine. Loopy uses your CLI agents and your keys instead of a hosted sandbox.

Everything in one loop

Shipping fails in the gaps. Loopy closes them.

Research, memory, compression, build, QA, and ops — wired deep, not bolted on.

Research dump

48k tokens

SuperCompress

Agent sees

12k tokens · 74% saved

Research without context blow-up.

Long Tavily and Firecrawl dumps get compressed before your coding agent ever sees them.

Codex
Claude Code
Cursor

Same loop · local CLI auth

Pick your coding agent.

Same loop — whichever local CLI you already use. Codex and Claude Code authenticate locally.

GitHub repo updated
Linear ticket synced
Ship summary sent

Ops when the build finishes.

Repo updates, tickets, deep research, and email summaries — automated after every run.

For years, solo founders hired one agent and hoped. The next wave runs a whole company — planner, researchers, builders, ops — on one machine. Loopy is that company.

The stack behind your agent company

Each integration is a specialist on the team — wired into a real pipeline phase, not a logo slide.

Tavily

Firecrawl

Mem0

SuperCompress

Nebius

Composio

GitHub

Linear

Gmail

Codex

Claude Code

Cursor

Stripe

Qdrant

MotherDuck

TavilyResearch & ideation

Mem0Persistent memory

SuperCompressContext eviction

NebiusCEO · plan · route · brand

ComposioGitHub · Linear · Gmail

FirecrawlDeep scrape

OpenClawLocal agent runtime

StripeSDK template

QdrantRAG SDK template

MotherDuckAnalytics SDK template

Local by design

Built to run on your machine

Privacy

Your keys, your machine

Loopy runs entirely locally. API keys live in ~/.loopy/ — never uploaded, never shared.

No hosted agent runtime
Configure everything via local UI
Open source — audit the code yourself

Portfolio

Run a portfolio of products

Multiple projects, concurrent builds, per-project turn limits. Nebius plans each build to finish inside its budget.

Up to N concurrent builds (configurable)
Per-project max turns + global default
Queue when all slots are full

Agents

Bring your own coding CLIs

Loopy orchestrates Codex, Claude Code, OpenCode, or Cursor — spawn parallel agents per turn when you need speed.

Same loop, any local CLI
Agent auth stays on your machine
Switch agents per project in the dashboard

Stack

Deep integrations, not bolt-ons

Tavily, Mem0, SuperCompress, and Composio are wired into every phase — research through ship.

Context compression before every build
Persistent memory across runs
GitHub, Linear, Gmail after every loop

Quickstart

One command. No clone, no source on disk — the installer drops a loopy command via pipx or an isolated env it manages for you.

# Install
curl -fsSL loopy.yachts/install | bash
loopy serve
# → http://127.0.0.1:8787 → Open Loopy → onboard → Loop

# Headless
loopy run --idea "CLI that syncs Linear to markdown" --agent codex
loopy run --verbose   # stream agent output
loopy run --demo
loopy doctor

# From source (for contributors)
git clone https://github.com/arjunkshah/loopy.git && cd loopy && ./bootstrap.sh

Questions & Answers

Is Loopy one agent or many?

A company of agents. Nebius plans and routes turns. Tavily and Firecrawl research. Mem0 remembers. Your coding CLIs build — optionally in parallel. Composio handles GitHub, Linear, and Gmail ops after each loop.

Can I run multiple projects at once?

Yes. Loopy supports multiple projects with a configurable concurrent build limit. Extra projects queue automatically until a slot opens.

What is the max turn limit?

Each project can have a turn budget (default 200). Nebius receives this from the start — it sizes the build plan and routes each turn to ship before the budget runs out. Set it globally in settings or per project when you start a loop.

Is Loopy hosted?

No. Loopy runs entirely on your machine. This website is documentation and overview only.

Where do API keys go?

~/.loopy/config.json — never uploaded anywhere. Configure via the local UI at loopy serve.

What is SuperCompress?

A learned memory layer that compresses research context before your coding agent runs. Vendored from the SuperCompress project.

Can I try without API keys?

Yes — loopy run --demo runs the full pipeline in demo mode with scaffold output.

Can I steer the agent mid-run?

Yes — Manual takeover lets you queue steering messages during an active build. Loopy injects them into the next agent turn so you can redirect without babysitting every step.

Does Loopy email me progress?

Connect Gmail via Composio and set your notify address. Loopy sends a kickoff email at start, a progress email after each turn, and a completion email when the build finishes.

How long can it run unattended?

Default turn budget is 200. Nebius plans within that from turn one — typical unattended runs last up to nine hours. Use the overnight queue and scheduler for portfolio builds.

Which coding agents are supported?

43 registered agents including Codex, Claude Code, Cursor, Gemini, Copilot, OpenClaw, and Freebuff — pick per project in the dashboard or pass --agent to the CLI. Run loopy doctor to see what's on your PATH.

Put your backlog on autopilot.

A team of agents that never stop working — governed, reviewable, and learning every run. All local, all yours.

or clone on GitHub →