locomo

A minimalist MVP demonstrating a simple yet profound insight: aligning AI memory with human episodic memory granularity. Shows how this single principle enables simple methods to rival complex memory frameworks for conversational tasks.

An open-source, vector-free long-term memory engine for AI agents, achieving SOTA on LoCoMo and LongMemEval with significantly less context.

Local-first AI memory — runs offline on any machine with 8 GB+ RAM (SBC, mini PC, laptop, workstation). Zero-loss verbatim archive, knowledge graph, hybrid retrieval. Framework-agnostic, no cloud.

Open-source causal graph memory for AI agents. 85.55 on LoCoMo under a frozen, reproducible protocol. MCP server with ACT-R scoring, spreading activation, and active forgetting.

Full audit of the LoCoMo benchmark

Open evaluation harness for AI agent memory systems. Runs LoCoMo and LongMemEval against Synap, Mem0, Zep and Supermemory with pluggable provider adapters.

Multi-agent memory substrate for PostgreSQL — provenance-gated, vector-hybrid recall

Auditable memory layer for AI agents: zero-LLM-call local ingest (~10ms/msg, air-gapped), matches Mem0 on accuracy at ~1000x lower ingest cost, bi-temporal belief-state, MCP server. Honest LoCoMo/LongMemEval benchmarks. Open source (Apache-2.0).

Research benchmark: conversational memory learning on a corrected LoCoMo dataset

A memory companion that lives in your AI workflow

🚀 A modern, production-ready refactor of the LoCoMo long-term memory benchmark.

Drop-in FastAPI proxy for llama.cpp, Ollama, vLLM and similar backends. Automatically prunes, summarizes, and extracts the most relevant context from large inputs using advanced strategies (readagent, rlm, embeddings) so your local models answer accurately without cloud services.

Benchmarking markdown knowledge graphs as agent memory — LOCOMO and HotPotQA tracks with Claude Code

A Beautiful portfolio site

Token-budgeted long-term memory for AI agents — extract once, select per question under a hard token budget. Pure-math retrieval, no LLM on the hot path.

Reproducible LoCoMo, LongMemEval, PersonaMem, FAMA, recall and latency benchmarks for long-term AI memory systems

Per-speaker memory isolation with neural reranking for multi-party LLM agents. 66.6% on LoCoMo (+50.2 pp on temporal over RAG) at 1.02 LLM calls per query.

Independent benchmark for memory ingest & retrieval: TiMEM vs MemOS Cloud vs Mem0 Platform (LoCoMo)

Reproduce Verbative's memory benchmarks — LoCoMo and AMA-Bench (all six domains) — against your own installed Verbative memory engine. Includes the harness, the judge-host calibration study, and the published per-question answers and verdicts behind every number we quote.