APP

Jeli

Security and governance layer for sovereign personal memory

hash-chained provenance, constitutional rules, judicial conflict resolution, portable memory

What it is

Jeli is a security and governance layer for personal AI memory systems. It keeps facts, decisions, preferences, history, and agent-written context under your control, with a verifiable chain of custody for every memory.

It is not a promise that memory can be made perfectly secure. Jeli makes poisoning and tampering harder, more visible, and easier to audit through configurable layers: cryptographic provenance, trust scoring, user-signed rules, conflict resolution, and exportable data.

What changed

The project has moved beyond a simple personal memory store. Jeli now implements a three-branch governance model for memory:

Executive: agents

Agents propose memories through a scoped MCP server. They cannot write directly to the database or claim user-level trust.

Legislative: storage

PostgreSQL and pgvector hold an append-only, HMAC hash-chained memory log with audit rows and state events.

Judicial: resolution

Daemons arbitrate contradictions, record precedent, and escalate unresolved conflicts for human review.

Constitutional: user rules

User-signed rules can block writes, cap trust, filter reads, and enforce non-negotiable memory policy.

Current capabilities

Why it matters

Memory is becoming one of the most important parts of AI infrastructure. If your assistant remembers your preferences, work, relationships, decisions, and identity, you need more than convenience. You need custody, provenance, revision history, conflict handling, and a real exit path.

Jeli is built so memory remains something you can inspect, move, verify, revise, invalidate, redact, and govern.

Status

v0.2.0-alpha. The three-branch governance model is implemented and tested, including the constitutional layer, judicial precedent, entity graph, portability, ingestion bouncer, and read/write defense layers. Jeli has been deployed on local hardware since v0.1.0-alpha.

See the GitHub repo for the latest status, threat model, and trust doctrine.