moorcheh-ai/memanto

What it solves

AI agents often suffer from context resets, where they forget previous instructions or facts once a session ends. Traditional memory tools are often passive, requiring agents to manually query and parse data, which consumes tokens and adds latency. Memanto solves these issues by providing an active memory layer that persists context across sessions, reduces token usage through relevant retrieval, and eliminates indexing delays.

How it works

Memanto acts as a companion agent that performs three core operations: remember, recall, and answer. It uses an information-theoretic semantic engine (Moorcheh) to provide exact retrieval without the need for traditional vector database pipelines or reranking. It categorizes memories into 13 built-in types (such as instruction, fact, and goal) and supports temporal queries to account for recency. Users can run it fully locally via Docker/Ollama or use a managed cloud service.

Who it’s for

It is designed for developers and users of AI coding assistants and agents (such as Claude Code, Cursor, Windsurf, and GitHub Copilot) who need their agents to maintain long-term continuity and avoid re-explaining codebases or preferences in every new session.

Highlights

• Active Memory Management: Performs remember, recall, and answer operations automatically.
• Zero Indexing Latency: Memories are searchable the instant they are stored.
• Typed Semantic Memory: Uses 13 categories to organize information by type (e.g., decisions, preferences, goals).
• High Portability: Supports an Open Knowledge Format (*.md files) to allow users to migrate memory estates between different tools.
• Broad Integration: Connects with over 20 agents including Cursor, Windsurf, and Cline via a simple CLI command.