Memro – Persistent Memory for Claude Code, Cursor and Codex

Memro has introduced an open-source memory infrastructure designed to give AI agents persistent memory across different sessions. The project includes a Go server, SDKs for Python, JavaScript/TypeScript and Go, a Model Context Protocol (MCP) server, and various integrations with popular coding assistants like Claude Code, Cursor and OpenAI Codex.

The project is licensed under PolyForm Noncommercial License 1.0.0, making it free for non-commercial use including personal study, experiments, hobby projects, research, and testing. Commercial use requires either a managed subscription on Memro Cloud or an enterprise commercial license.

Deployment Options

Users can deploy Memro in two ways: through the Memro Cloud service or by self-hosting. For cloud deployment, users can connect via device login using the command 'npx memro-cli connect' and check status with 'npx memro-cli status'. The self-hosted option uses Docker Compose to run the full Memro API server and PostgreSQL with pgvector on local infrastructure.

The self-hosted setup can be implemented either by cloning the repository and running 'docker compose up -d' or by using prebuilt images with a one-liner command. The backend API runs on port 8081, the dashboard on port 5173, and PostgreSQL on port 55432 to avoid collisions with existing local setups.

Technical Architecture

Memro operates on a Go HTTP API backed by PostgreSQL and pgvector. The system combines vector cosine similarity and full-text keyword search using Reciprocal Rank Fusion (RRF, k=60). Agents authenticate using sovereign Ed25519 keypairs solved via local proof-of-work or standard Bearer API tokens.

The project includes advanced memory features beyond basic vector search, including typed memories (semantic, episodic, procedural, profile), temporal search with timestamp weighting, conflict analysis and resolution, and reasoning graphs that trace memory provenance and dependencies.

SDK Integration

Memro provides official SDKs for multiple programming languages. The Python SDK allows users to create an agent with 'Agent.create()', remember information with 'agent.remember()', and recall memories with 'agent.recall()'. Similar functionality is available in JavaScript/TypeScript and Go SDKs, each with their own syntax and implementation details.

The project also includes integration guides for connecting with popular coding tools and has published empirical evaluations including the LongMemEval Benchmark, which tests multi-session recall and knowledge updates across 500 questions. Contributing to the project requires setting up a PostgreSQL database with pgvector and running the backend API server locally.