Skip to content
@reticlehq

Reticle

Reticle



We build the proof layer for AI coding agents.

An agent can write the code. It cannot tell you whether the code works — reading a diff proves nothing, unit tests don't run the app, and a screenshot can't see a silent 500. So agents say "done" over broken software, and you find out by opening the app.

We're closing that gap: giving agents runtime perception of a real running application — the network calls, the store, the console, the routes, the DOM — and a verdict they can act on. Pass, fail, or "couldn't tell", with the file:line to fix.

reticle.sh · Docs · Discord · Main repo →


The bet

Everything else that checks a web app looks at it from the outside — pixels, DOM, a vision model guessing at a screenshot. Reticle looks at the program: app state, framework signals, request cardinality, swallowed errors. That's the bug class an outside-in tool structurally cannot catch, and it's where AI-written code actually fails.

A dev-only SDK runs inside your app. A local bridge exposes it to your agent over MCP. The agent drives the real app, reads what actually happened, and fixes it before you ever see it.

No screenshots. No vision model. No test syntax. Runs on your machine, on localhost, and your app data never leaves it.

📌 Pinned

The whole thing — browser SDK, MCP server, CLI, framework adapters, spec runner. Apache-2.0.

typescript mcp ai-agents verification

stars npm

Getting started, the tool surface, token efficiency — and llms.txt + an MCP endpoint, so your agent reads them directly.

docs llms-txt mcp

site

Works with Claude Code, Cursor, Copilot, Codex, Windsurf, OpenCode, and any MCP agent. React, Vue, Svelte, Preact, Astro, plain HTML · Next.js · Electron & Tauri · macOS, Linux, Windows.

/plugin marketplace add reticlehq/reticle     # Claude Code
npx skills add reticlehq/reticle              # everywhere else

Where this goes

  • Verification the agent can't fake. A green verdict never rests on evidence Reticle doesn't have. Every false-green we find gets fixed and pinned with a regression guard — that invariant outranks every feature.
  • Diagnosis, not just detection. First divergence, blast radius, and the exact file:line — because the win we measure is fewer agent tool-calls to a fix, not a longer bug list.
  • Zero-install on-ramp. Drive an app with no SDK wired at all, and earn the deeper read.
  • Beyond the browser. React Native, native Android, iOS — being explored honestly, and not promised until an agent can drive a real app there and get a verdict.

The core verification engine is free forever. Full direction lives in the Roadmap.

Build it with us

The work happens in the open — what's being built this cycle, what's up for grabs, and design calls before they land.

💬 Join the Discord → · Good first issues · Contributing


Discord X Website npm

dev-only · localhost-only · your app data stays local · Apache-2.0

Popular repositories Loading

  1. reticle reticle Public

    AI agents can generate code, but they still struggle to understand what they build. Reticle gives them runtime perception of web & desktop applications.

    TypeScript 344 81

  2. .github .github Public

    Organization profile and shared community health files

Repositories

Showing 2 of 2 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…