Why Mneno?
Long-running AI systems accumulate stale facts, duplicated notes, noisy transcripts, and vague summaries. That context eventually makes retrieval worse, not better. Mneno is built around scoring, lifecycle management, compaction, and explainable decisions so you can see what memory was retrieved, excluded, promoted, merged, discarded, or superseded and why.Key Features
- Memory Scoring: Rank memories by recency, relevance, importance, and freshness signals.
- Explainable Retrieval: Search returns ranked results with score components and reasons.
- Auto-Compaction: Preserve key claims when context gets too large or noisy.
- Context Building: Build deterministic context packages for prompt injection within a token budget.
- Persistent Storage: Support for in-memory, JSON file, and SQLite storage.
- Provider Agnostic: Protocol-based integration for embeddings, LLMs, and rerankers.
- Lifecycle and Hierarchy: Preserve audit history across active, superseded, archived, and conflicted memories organized into deterministic layers.
- Sessions and Timelines: Group memories into ordered sessions, reconstruct timelines, and recover continuity across related work.
- Structured Extraction: Extract memories deterministically or through an optional validated
LLMProvider. - Local Observability: Record in-memory traces for retrieval, context building, compaction, conflicts, hierarchy, extraction, and sessions.
- Evaluation Hooks: Measure retrieval, context, compaction, token efficiency, latency, and explainability for future Mneno Bench adapters.
Getting Started
Quickstart
Get Mneno up and running in minutes.
Core Concepts
Learn how Mneno handles memory and context.
Observability
Debug memory decisions with local traces.
Evaluation
Produce benchmark-friendly metrics and exports.