Reconciling memory for AI agents.
Most agent memory accumulates — it saves every fact and later hands your agent a stale one. Vayl reconciles: a new value supersedes the old, "we dropped X" actually retracts X, and you can still ask what was true before. It speaks the Model Context Protocol, so any MCP client (Claude Desktop, Cursor, Claude Code) plugs in.
"We use Redux." → remembered
"Actually we moved off Redux to Zustand." → Redux retired, Zustand active
"What do we use?" → "Zustand" (not "Redux, Zustand")
"What did we use first?" → "Redux" (history kept)
- ✅ Reconciles, doesn't pile up — one live value per fact, so "what's true now" is unambiguous.
- 🗑️ Removal is first-class — retractions actually remove; graph stores can't model this, additive stores never do.
- 🕰️ Auditable — every change is on a tamper-evident, Ed25519-signed audit chain.
- 🔒 Local & private — SQLite by default (optional Postgres), encrypted at rest, no telemetry.
📚 Documentation · 📦 PyPI
pip install vayl-mcp| Repo | What |
|---|---|
| vayl | The reconciling-memory engine + MCP server. Two core dependencies, offline-testable. |
Apache-2.0 · maintained by @ac12644