- Tavily search
- Firecrawl scrape
- SuperCompress
- Codex build
65%KV saved per run
Research without context blow-up.
Long Tavily + Firecrawl dumps get compressed before your coding agent ever sees them.
Local, open source, your keys. Tavily finds the idea. Firecrawl deep-researches. SuperCompress keeps your coding agent under budget. Codex, Claude Code, OpenCode, or Cursor Agent builds it. Composio runs GitHub, Linear, Firecrawl, and Gmail.
Run loop
MVP scaffolded in ~/.loopy/projects/a3f2 — Linear ticket created, Firecrawl research cached, Gmail summary sent.
65%KV saved per run
Long Tavily + Firecrawl dumps get compressed before your coding agent ever sees them.
Same loop — whichever local CLI you already use. Codex and Claude Code authenticate locally.
Repo updates, tickets, deep research, and email summaries — automated after every run.
1 · Ideate
Web search + synthesis. Optional seed topic — or let Loopy pick from market signals.
2 · Research
Top URLs become build context — docs, competitors, references.
3 · Compress
Learned eviction keeps token usage stable so Codex sessions stay affordable.
4 · Build
Local CLI — Codex and Claude Code use their own login; OpenCode and Cursor when installed.
5 · Operate
GitHub, Linear, Firecrawl, Gmail — updated after every run.
Local UI
loopy serve at localhost — keys in ~/.loopy/ only.
Install any CLI on your PATH. Pick per project in the local dashboard.
OpenAI Codex CLI — codex exec --full-auto. Authenticates via the local Codex login.
Anthropic's agent CLI. Authenticates via the local Claude Code login.
Open-source coding agent. Runs from a generated prompt file in your project dir.
Cursor's local agent CLI when installed — same loop, your subscription.
Clone, install, onboard, loop.
# Install git clone https://github.com/arjunkshah/loopy.git cd loopy ./bootstrap.sh source .venv/bin/activate loopy serve # → http://127.0.0.1:8787 → Open Loopy → onboard → Loop # Headless loopy run --idea "CLI that syncs Linear to markdown" --agent claude loopy run --demo loopy doctor
No. Loopy runs entirely on your machine. This website is documentation and overview only.
~/.loopy/config.json — never uploaded anywhere. Configure via the local UI at loopy serve.
A learned memory layer that compresses research context before your coding agent runs. Vendored from the SuperCompress project.
Yes — loopy run --demo runs the full pipeline in demo mode with scaffold output.