MEMANTO is an active memory agent designed to provide AI agents with persistent, queryable context across sessions, addressing limitations of passive memory tools. It targets developers building sophisticated AI agents, enabling them to achieve long-term goals and avoid confusion through state-of-the-art retrieval and zero ingestion latency.

How It Works

MEMANTO functions as an active agent, offering three core operations: remember, recall, and answer. It tackles six identified gaps in agent memory, including static injection, temporal decay, lack of provenance, flat memory structures, no writeback capabilities, and indexing delays. The system leverages a typed semantic memory schema combined with Moorcheh's information-theoretic retrieval engine, a no-indexing semantic database. This architecture enables exact search, sub-90ms retrieval times, and zero ingestion latency, avoiding the complexity of graph databases and LLM extraction bottlenecks.

Quick Start & Requirements

• Install: pip install memanto
• Configure: Run memanto in your terminal; it prompts for your Moorcheh API key. Native LLM access is included, negating the need for separate external model API keys for common workflows.
• Documentation: Official docs are available at https://docs.memanto.ai. Setup guides are also on the Moorcheh YouTube channel.

Highlighted Details

• Achieves state-of-the-art benchmarks: 89.8% on LongMemEval and 87.1% on LoCoMo, outperforming competitors like Mem0 and Zep.
• Supports 13 built-in memory types (e.g., instruction, fact, decision, preference) for structured storage and retrieval.
• Offers zero ingestion latency; memories are searchable instantly upon storage.
• Integrates with numerous multi-agent ecosystems, including Claude Code, Codex, Cursor, GitHub Copilot, and others.
• The underlying Moorcheh database provides instant write-to-search, exact search accuracy, and zero idle costs via a serverless architecture.

Maintenance & Community

Support and community interaction are available via their Discord server. Additional resources and setup guides can be found on the Moorcheh YouTube channel.

Licensing & Compatibility

MEMANTO is released under the MIT License, which is permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

A Moorcheh API key is required for operation. MEMANTO does not currently offer a hosted API server; users must run their own local server using memanto serve or Docker, which may represent an initial setup hurdle.