I built this after getting frustrated with Claude Code forgetting everything between sessions.
*The problem:* Every new session, AI has no memory. You end up re-explaining architecture, past decisions, why you chose X over Y. Gets old fast.
*What CodeSyncer does:* Records AI decisions as comment tags directly in your code.
```typescript // @codesyncer-decision [2026-01-15] Chose sync over async (UX feedback) // @codesyncer-inference Min $1 (Stripe policy) async function processPayment(amount: number) { ... } ```
Next session, AI reads the code and instantly knows the context. No re-explaining.
*Setup:* ```bash npx codesyncer init npx codesyncer watch ```
*Why not just use CLAUDE.md?* CLAUDE.md is great for rules ("use TypeScript strict mode"). But it doesn't capture why specific code was written a certain way. CodeSyncer stores that context where it belongs—in the code itself.
*Similar project:* Steve Yegge's Gas Town (https://github.com/steveyegge/gastown) solves this for teams with 20-30 agents. CodeSyncer is the solo developer version—simpler, 5-min setup.
Would love feedback, especially: - Does this solve a real problem for you? - What's missing? - Any concerns with the approach?