# Zengram

A shared memory system that lets multiple AI agents, such as Claude Code running on different machines, store and recall facts, decisions and status through one Postgres-backed store.

Zengram gives separate AI agent sessions, for example Claude Code on a laptop and an autonomous agent on a server, access to the same memory, so something one agent learns can be recalled by another without re-explaining it. Memory is typed: events are immutable history, facts upsert by key so a new fact supersedes an old one, statuses track current state, and decisions record a choice with its reasoning, each with its own decay and mutation rules.

Everything is stored in one Postgres database using pgvector for semantic search alongside structured tables for exact queries, and a search runs up to three retrieval paths in parallel, vector similarity, full-text search and an optional entity-graph expansion, fused with weighted reciprocal rank fusion. An optional cross-encoder reranking step can then re-score the fused results, with the project reporting roughly double the ranking quality on its own bilingual production data when enabled.

Website: https://zensystem.ai