MembraneSelective Learning and Memory for Agentic Systems

A structured memory substrate that lets AI agents remember, learn, and forget -- just like biological cognition.

Five Memory Types

Episodic, working, semantic, competence, and plan graph memories mirror how humans store and recall different kinds of knowledge.

Trust-Aware Retrieval

Every query carries a trust context that gates access by sensitivity level and scope, keeping private data private.

Automatic Consolidation

Raw experiences are periodically distilled into stable facts, learned procedures, and reusable plans -- no user intervention required.

Decay and Reinforcement

Salience decays over time using configurable curves. Frequently accessed memories are reinforced; stale ones fade away.

Revisable Knowledge

Semantic facts support supersede, fork, retract, and merge operations with full audit trails.

gRPC API

A 15-method gRPC service exposes ingestion, retrieval, revision, reinforcement, and metrics to any language.

Quick Start

bash

# Build from source
git clone https://github.com/GustyCube/membrane.git
cd membrane
make build

# Start the daemon
./bin/membraned -db membrane.db -addr :9090

See the Getting Started guide for a complete walkthrough.