Until now.
Cursor, Claude Code, Copilot. They all read your repo for context. The question isn't whether AI will read your repo. It's whether it'll find specs, or nothing.
Edit markdown in GitHub's textarea. Pray nobody else is editing the same file.
Write docs in Google Docs. Copy-paste into a PR. Lose all formatting.
Ask the new engineer to write the spec. They've never seen the codebase.
It understands your repo, your team, and your workflow.
The application uses a microservices architecture with event-driven communication via Redis Streams.
No codebase download. No copy-paste from Google Docs. No Jira ticket asking someone else to do it.
Should we mention the rate limiting here?
Added the deployment section you requested.
Looks great! Let's move this to review.
Live cursors & presence. See who's editing, where, in real-time.
Conflict-free editing. Built on Yjs CRDTs. No merge conflicts. Ever.
Threaded comments. Pin feedback to specific lines. Resolve when addressed.
Editor & viewer roles. Share drafts with the right level of access.
Codebase-aware context. AI reads your actual source files.
Slash commands: /outline, /summarize, /fix, /translate, /tone, /expand
Ghost text autocomplete. Copilot-style suggestions as you type.
Edit-in-place. AI proposes diffs. Accept or revert with one click.
Your keys, your models. Bring Claude or OpenAI. Keys never leave the server.
Specs become prompts. .md files in your repo are context for every AI tool.
# API Reference ## Endpoints ### GET /api/users Returns a list of users...
# ADR-001: Use Event Sourcing ## Context The system needs to track all state changes...
# Welcome to the Team ## First Day Setup 1. Clone the repository 2. Install dependencies...
# v2.1.0 Release Notes ## What's New - Real-time collaboration - AI-powered editing...