vibe-brain

# 🧠 Vibe Brain v3.2

**Your AI coding agent never forgets. Not between sessions. Not mid-conversation.**

Persistent project memory that gives your AI assistant full context about your codebase — architecture, features, APIs, database schema, decisions, active threads, and exactly where you left off — on every single conversation. Smart loading saves tokens. Mid-session refresh prevents context decay. Thread tracking gives continuity across sessions.

Works with **Claude Code** · **Cursor** · **Cursor** · **Windsurf** · **Cline** · **GitHub Copilot** · **Antigravity** · **Any AI coding agent**

---

## The Problem

Every new conversation:
> *"This is a Next.js app with Prisma..."*
> *"We use tRPC, not REST..."*
> *"We already built auth last week..."*

And mid-conversation (prompt 15+):
> *"You're forgetting what I said earlier..."*
> *"We already decided on Ably, not Pusher..."*
> *"Stay focused on what I asked..."*

You waste tokens re-syncing. Past decisions get forgotten. Features get re-proposed. Context degrades after ~10 prompts. You go in circles.

**Vibe Brain fixes all of this. Permanently.**

---

## How It Works

```
┌───────────────────────────────────────────────────────────────┐
│                       YOUR PROJECT                            │
│                                                               │
│  BRAIN.md ← Master blueprint. Smart-loaded.                  │
│  ↑          Core sections on start. Rest on-demand.           │
│  │                                                            │
│  │ auto-compress         ← every 50 lines                    │
│  │ health check          ← every 5th compression              │
│  │ context refresh       ← every 10 prompts                  │
│  │                                                            │
│  SESSION.md ← Live buffer + active threads.                   │
│               Logs actions. Tracks work topics.               │
│               Rese