✨ Now in Beta — The spec layer for AI coding

What if specs and AI agents were friends?

AI coding, with your team in the loop. Write specs together, review AI plans, and execute with confidence.

See How It Works

AI gets involved too late

You write specs in Notion. Discuss in Slack. Track in Linear. Then copy-paste context into Claude.

By the time AI sees your code, it's missed all the decisions, constraints, and context that matter.

What if AI was in the room from day one?

feature-user-auth.md — Larry📄 User Authentication📄 API Endpoints📄 Database Schema📄 Error HandlingCOLLABORATORSAMSUser AuthenticationRequirementsAlexSarah2✨ AI SuggestionGenerate Plan →
“
A better way to use those tools, not a replacement for those tools.

Works with Claude Code, Cursor, Copilot

How It Works

From messy idea to merged PR in three steps

Step 1

Draft Together

Write specs collaboratively in real-time. Comments, suggestions, @mentions. Everyone aligned before code begins.

Step 2

Review AI Plans

AI generates step-by-step plans from your specs. Review each step, approve or reject, add comments before building.

Step 3

Execute & Ship

Watch Claude Code bring your plan to life. Isolated worktrees, real diffs, full audit trail. Merge when ready.

Plan Review — User Authentication3/5 approved✓Create users table with email, password_hash, created_atDatabase migration • Approved by AlexApproved✓Implement bcrypt password hashing utilitySecurity • Approved by SarahApproved✓Create /api/auth/register endpointAPI Route • Approved by AlexApproved4Create /api/auth/login endpoint with JWTAPI Route • Waiting for reviewApproveRejectComment5Add session management and logout...Pending
Plan Review

Approve every step before it ships

AI generates implementation plans from your specs. Review each step, approve what looks good, reject what doesn't. No surprises in production.

  • Step-by-step approval workflow
  • Comments and feedback on each step
  • Full audit trail of decisions
Execution

Watch your plan come to life

Claude Code executes your approved plan in an isolated worktree. See real-time progress, review every diff, merge when you're ready.

  • Isolated git worktrees per execution
  • Real-time terminal output
  • Full diff review before merge
Execution — User AuthenticationRunningTerminal — Claude Code$ claude-code execute plan.json✓ Step 1: Create users table✓ Step 2: Password hashing✓ Step 3: Register endpoint→ Step 4: Login endpoint Creating src/api/auth/login.ts Writing JWT token logic...Progress: 4/5 stepsChanges — src/api/auth/login.tslogin.ts +48schema.ts +12utils.ts +812345678+ import { sign } from 'jsonwebtoken'+ import { compare } from 'bcrypt' + export async function POST(req) {+ const { email, password } = await+ const user = await db.query...+ const valid = await compare(...)+ const token = sign({ userId }...)3 files changed+68 additions-4 deletions✓ Merge ChangesView Full Diff

Built for teams that ship

Real-time Collaboration

Multiple cursors, live sync, presence indicators. Like Google Docs for specs.

Plan Review

Approve, reject, or comment on each step before code is written.

Isolated Worktrees

Each execution runs in its own git worktree. No dirty main branch.

BYO Execution

Works with Claude Code, Cursor, and Copilot. We orchestrate, you execute.

AI in the Room

AI analyzes specs in real-time, suggests improvements, catches issues early.

Full Audit Trail

Every decision tracked. Who approved what, when, and why.

AI Code Review

AI surfaces issues and suggestions in your diffs before you merge.

Git Integration

Commit, push, and open PRs with AI-generated commit messages from your diffs.

Board View

Kanban-style spec board. See every feature's status at a glance.

Powered by Memable

Project Memory

Larry remembers what it learns about your project — patterns, decisions, constraints — and brings that context into every plan and conversation. The longer you use it, the better it understands your codebase.

Simple, transparent pricing

Start free. Bring your own Claude Code subscription for execution.

Free

$0/month

Perfect for trying out Larry

  • 5 plan generations/month
  • 3 executions/month
  • 3 projects
  • 3 collaborators per spec
  • Community support
Get Started Free
Most Popular

Pro

$25/month

For individual developers

  • Unlimited plan generations
  • 50 executions/month
  • Unlimited projects
  • 10 collaborators per spec
  • Priority support

Team

$50/month

For growing teams

  • Unlimited plan generations
  • 200 executions/month
  • Unlimited projects
  • 50 collaborators per spec
  • Audit logssoon
  • Dedicated support
Start Free Trial

Enterprise

Unlimited usage (BYO API key), self-hosted option, custom integrations, 99.9% SLA

Contact Us

Frequently asked questions

Do I need to install anything?
Yes — a small local daemon handles AI execution on your machine. It's a one-time setup and the UI walks you through it step by step. You can start writing and collaborating on specs immediately without it; the daemon is only needed when you're ready to generate plans or execute code.
Is my code private?
Your code never leaves your machine. The local daemon runs entirely on your computer and only sends specs (not your codebase) to the cloud. Plan generation and code execution happen locally using your own Claude Code subscription.
What AI subscription do I need?
You bring your own Claude Code subscription. Larry uses it for plan generation and execution — no additional AI costs on top of your subscription. We orchestrate; you supply the compute.
Can I use this solo or do I need a team?
Works great solo — many developers use Larry to structure their own AI-assisted workflow before looping in teammates. Team features like shared specs, @mentions, and collaborative plan review kick in whenever you're ready to add collaborators.
What's the difference between a generation and an execution?
A generation creates the step-by-step plan from your spec. An execution runs that approved plan in an isolated git worktree. You can generate multiple plans and only execute the one your team has reviewed and approved.
Does it work with tools other than Claude Code?
Claude Code is the primary execution engine today. Support for Cursor and Copilot is on the roadmap. The collaboration, spec writing, and plan review workflow works independently of which AI tool you use for execution.

Ready to bring AI into the room?

Start collaborating for free. No credit card required.
Bring your own Claude Code subscription for execution.

🚧 STAGING ENVIRONMENT 🚧