Mnemo Cortex: Open-Source Memory Coprocessor for AI Agents

Mnemo Cortex is an open-source memory coprocessor designed for AI agents, offering persistent recall, semantic search, and crash-safe memory capture without requiring hooks. It includes features like cross-agent overnight synthesis, an auto-compiled knowledge base (WikAI), agent-to-agent messaging, and a structured facts store with confidence tracking.

RDR87Confidence 90%AI agentsmemoryopen-sourcesemantic searchknowledge basecontext managementPythonLLM

Why it matters

Mnemo Cortex addresses a critical challenge in AI agent development: persistent memory and context management. By providing a robust, open-source solution for deep recall, cross-agent synthesis, and structured fact storage, it enables more capable and consistent AI agents. The recent v2.12.0 update introduces structured facts with confidence tracking and an embedding hosted fallback, enhancing reliability and functionality for developers building complex agent systems.

Mnemo Cortex is an open-source memory coprocessor for AI agents, developed by GuyMannDude. It provides persistent memory across sessions, enabling semantic search and crash-safe capture without requiring direct code hooks. The project is written in Python and is licensed under MIT.

Key features of Mnemo Cortex include:

* **Deep Recall:** Persistent memory and semantic search capabilities. * **Dreaming:** A cross-agent overnight synthesis process that compacts individual agent memories into themes and merges them into shared context, allowing agents to wake up with knowledge of others' activities. This is highlighted as a unique feature among AI memory systems. * **WikAI:** An auto-compiled knowledge base that regenerates nightly from Mnemo, ensuring up-to-date information. * **Sparks Bus:** An agent-to-agent messaging system with delivery confirmation. * **Structured Facts:** A key-value store designed for exact lookups of entity attributes, names, and settings, complementing semantic search. It includes a three-state confidence ladder (verified, high_probability, false) and an audit log for changes.

The latest release, v2.12.0, introduces several significant updates:

* **Phase 3 Structured Facts Table + Three-State Confidence:** A new SQLite store for structured facts with a composite primary key, an append-only audit log, three-state confidence tracking, and new HTTP routes and MCP bridge tools for fact management. It also includes an optional contradiction notification system via bus messages or Discord webhooks. * **Embedding Hosted Fallback:** Integration with Google Matryoshka for `GoogleEmbedding.embed()`, allowing for a hosted fallback when local Ollama embeddings are unavailable, ensuring system resilience. * **Dreamer Pipeline Rehab:** Fixes for issues in the `mnemo-dream.py` pipeline, including path migration, agent auto-discovery, and a two-stage map-reduce synthesis process with bounded token usage for predictable costs.

Mnemo Cortex aims to provide fluid memory and deep recall for various AI agent setups, with non-interactive installation and compatibility with platforms like Claude Desktop, LM Studio, AnythingLLM, OpenClaw, Agent Zero, and Ollama.