# LemonCrow Runtime

A local review workspace for AI coding agent output that pairs a faster agent runtime with code intelligence, exact-range reads and verification evidence, so a human can review revisions rather than another AI-written summary.

LemonCrow is a local review workspace backed by code intelligence, exact-range reads, bounded tool output, memory, routing and verification, built around the idea that human review state, not another generated review summary, should be the durable record of what an AI coding agent changed. It reports impact, provenance and verification evidence for each change, and preserves reviewed work across revisions so only new or changed work returns to the review queue.

It claims roughly 30 percent lower cost and 25 percent faster completion on average without sacrificing task quality across supported coding-agent hosts. A hosted install runs a thin client plus host integrations against a remote LemonCrow server; a local production install builds a loopback server into the bundle on port 7420. The project explicitly warns against placing a reverse proxy or tunnel in front of that local port to expose it remotely, and directs remote review needs to the hosted path with its own authentication.

Once installed, a session-start hook opens a workspace view automatically; review can also run from the terminal. Anonymous aggregate telemetry is on by default and can be disabled.

Website: https://lemoncrow.com