This is an early release preview. You may encounter bugs.
Amnesia logo

Tool memory

Amnesia

Memory layer for AI assistants and agents that persists user, session and agent context and retrieves it with multi-signal semantic search

C 63/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
0% recommend · 1 rater

01 / About

What Amnesia is.

Mem0 is a memory layer that gives AI assistants and agents persistent, personalized context: it retains user, session, and agent-level state, recalls preferences and history across conversations, and continues learning over time. It is available as an open-source library (Python and TypeScript) with a fully managed hosted option, and the README highlights use cases in customer support, AI assistants, healthcare, and adaptive productivity tools. Agents can even self-provision an API key from the CLI, with a human claiming the account later.

A revised memory algorithm (April 2026) replaced update/delete logic with single-pass ADD-only extraction — memories accumulate rather than being overwritten — added first-class storage of agent-confirmed facts, entity linking across memories, fused multi-signal retrieval (semantic, BM25, and entity matching), and time-aware ranking for queries about current state versus past events. Reported benchmark results, all single-pass retrieval:

Benchmark Old New Tokens Latency p50
LoCoMo 71.4 91.6 7.0K 0.88s
LongMemEval 67.8 94.8 6.8K 1.09s
BEAM (1M) 64.1 6.7K 1.00s
BEAM (10M) 48.6 6.9K 1.05s

The evaluation framework is open-sourced so the numbers can be reproduced independently.

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

03 / Build

Build with Amnesia.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Amnesia →

Browse the catalogue