YourMemory provides persistent, self-improving memory for AI agents, addressing the common problem of agents forgetting context and preferences between sessions. It targets developers and users of AI agents, offering a solution that mimics human memory processes to enhance recall and reduce repetitive interactions, moving beyond simple vector storage.

How It Works

YourMemory treats memory as a dynamic system rather than static storage. It employs consolidation to compress clusters of related facts into concise summaries, archiving originals to prevent data bloat. Biological decay, modeled on the Ebbinghaus forgetting curve, gradually fades unused memories while prioritizing frequently accessed ones. An entity graph links memories by shared concepts, enabling recall of relevant, even unprompted, information. This approach facilitates sharper, more contextually aware memory over time.

Quick Start & Requirements

Installation is straightforward via pip install yourmemory. It requires Python 3.11–3.14. The default setup uses DuckDB for local, zero-setup storage. For teams or production, PostgreSQL with the pgvector extension is supported via a DATABASE_URL. Optional local fact extraction can be enabled by installing Ollama (e.g., qwen2.5:7b, ~4.7 GB) or by setting YOURMEMORY_EXTRACT_BACKEND=anthropic. Standalone binaries (~2 GB) are available for offline use, bundling all dependencies. Setup involves registering a token from yourmemoryai.xyz and running yourmemory-setup. Official links include a live interactive demo and the project website.

Highlighted Details

• Performance Benchmarks: Claims +16pp better recall than Mem0 on LoCoMo and 2x better than Zep Cloud, with reproducible benchmark code available in the repository.
• MCP-Native & Local-First: Operates entirely on the user's machine without requiring API keys for core functionality, compatible with various Model Context Protocol (MCP) clients.
• Tamper-Evident Audit Trail: A hash-chained ledger logs all memory operations, allowing cryptographic verification of data integrity.
• Team Memory Pools: Enables role-based shared memory across teams while maintaining private memory isolation for individual users.
• Data Rights & Compliance: Provides one-command export and purge functionalities, aligning with SOC 2 controls for data access and erasure.
• Offline Querying: Capable of answering certain questions directly from memory without LLM calls, reducing latency and cost.

Maintenance & Community

The README does not detail specific maintenance contributors, sponsorships, or community channels like Discord or Slack. GitHub stars are provided as a social metric.

Licensing & Compatibility

The project is licensed under CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International). This license permits free use for personal, educational, academic research, and open-source projects. Commercial use necessitates a separate written agreement.

Limitations & Caveats

The CC BY-NC 4.0 license is a significant adoption blocker for commercial applications, requiring a separate licensing agreement. While installation is straightforward, advanced configurations involving external databases or local LLMs add complexity. Potential write lock contention issues exist when running both the MCP server and HTTP server concurrently with DuckDB, which can be mitigated by switching to SQLite or restarting services.