2
Show HN: Tandem – Real-time collaborative editor with AI attribution tracking
I built Tandem to solve a problem I kept running into with Claude Code: *How do you collaborate on documents with AI while maintaining proper attribution?*
Current tools (Google Docs, Notion, etc.) were designed for human-to-human collaboration. When I copy-paste Claude's suggestions into a doc, all attribution is lost. My team can't tell which parts I wrote vs AI-generated. In open source, this creates trust issues.
*Tandem's approach:* - Every edit is tagged as Human or AI - Git-based version control (full history, not just "last modified") - Real-time collaboration using Yjs CRDT - MCP integration (AI can directly edit documents using tools)
*Tech stack:* React 19, TipTap, Yjs, Hono, Bun, MCP
*Live demo:* https://tandem.irisgo.xyz
Think "Google Docs meets Git" but designed for human-AI teams. I'm treating this as a bottom-up approach to building an AI-native workspace (inspired by Sam Altman's recent comments about needing an "AI-native Slack").
Currently in MVP stage. Looking for feedback from the HN community: - Does this solve a problem you have? - What features would make this indispensable? - Concerns about AI attribution?
Happy to answer technical questions about the implementation!
*Why I built this:* Working with Claude Code transformed my coding workflow, but document collaboration was still broken. The copy-paste dance between AI and docs loses all attribution. In teams, this creates trust issues. In open source, it looks like plagiarism.
*Technical details:*
*Architecture:* - Frontend: React 19 + TipTap (collaborative editor) - Sync: Yjs CRDT over WebSocket - Backend: Hono on Bun (3x faster than Node) - Storage: Git for version control - AI: MCP protocol for tool integration
*The key innovation* is treating AI as a first-class participant in Yjs's CRDT. Each client (human or AI) gets a unique `clientID`, allowing per-character attribution.