Synthetic Memory that rhymes with you.

Synm is a user-owned memory layer for AI — portable context that lives on your hardware and is shared on your terms.

View on GitHub Learn more

What is Synthetic Memory?

Synthetic Memory is persistent, context-aware recall for AI agents that you own and control. It’s engineered, portable, and vendor-neutral — allowing continuity of knowledge across models and services while keeping data private.

Think of it as synonyms for your thoughts: similar memories retrieved when relevant, not a data dump. Synm uses local embeddings and policy-scoped sharing to provide just enough context, only when you permit it.

How it works

  1. Vault. Your notes and history live locally, indexed by embeddings.
  2. Mediator. A small API enforces profiles, scopes, TTLs, and redaction.
  3. Provision. External AIs request a scoped, temporary context when needed.
  4. Audit. Every disclosure is logged; you can revoke at any time.

Flow

User → AI tool → Synm API → (policy + redact) → Scoped context → AI tool
                                 ↳ audit log

Local first. Minimal disclosure. Your rules.

Ethos

Your mind, your memory, your rules. Synm is built to reverse platform lock-in. Models borrow context; they don’t keep it. Trust comes from architecture, not promises.

  • Privacy-first
  • Interoperable by design
  • Portable across vendors
  • Human-scale UX

Start building

Clone the reference implementation and run the Mediator + Vault stack locally.

Open the repo Contact

© 2025-09-07 Synm. MIT-licensed reference. The name “Synthetic Memory” may be used to describe compatible systems.