I gave ChatGPT persistent memory that actually works — open source

Enhancing AI Capabilities with a Persistent Knowledge Base: An Open-Source Solution

In the rapidly evolving field of artificial intelligence, enabling models to retain context across sessions remains a significant challenge. Traditional AI interactions often forget previous conversations or learned information, limiting their utility for complex, long-term tasks. Addressing this gap, a new open-source solution provides a robust, self-sustaining knowledge base system that grants AI agents persistent memory — all running on your own hardware.

Introducing a Custom Knowledge Base Server

The core of this project is a custom-built knowledge base server designed to store and organize a wide array of data, including articles, notes, code documentation, transcripts, tweets, and more. Users can feed their preferred content into the system, which then indexes all material into a lightweight SQLite database with full-text search capabilities. This index allows AI agents to query relevant information efficiently, providing context before generating responses. This approach minimizes hallucinations and enhances the factual grounding of AI outputs.

Seamless Integration with AI Agents

The knowledge base connects to various AI models through a custom connector utilizing the Model Context Protocol (MCP). This protocol standardizes communication, enabling multiple AI agents—such as ChatGPT, Claude Code, Codex, and Gemini—to access the shared knowledge repository. When you teach or update information within the system, all connected agents learn simultaneously, creating a unified, multi-agent ‘brain.’

Adaptive Learning Across Sessions

One distinguishing feature of this solution is its self-learning capability. Unlike simple retrieval-augmented generation (RAG) systems, this setup allows AI agents to remember what strategies and solutions worked in previous sessions. With each interaction, the system captures session data, including bug fixes and their contexts. Over time, this process leads to more efficient problem-solving—reducing the number of back-and-forth interactions needed to resolve issues, as the AI recognizes patterns and solutions based on accumulated experience.

Human Curation and Data Ingestion

To facilitate manual curation, the system integrates with Obsidian, a popular knowledge management tool. Users can clip articles, write notes, and bookmark valuable resources within their vaults. The knowledge base server then ingests this curated data, making it readily searchable by AI. However, the system does not require Obsidian; it can ingest any folder or markdown files, offering flexibility for various workflows.

Comparison with Industry Solutions

Recently, Google open-sourced their “Always-On Memory Agent,” which provides persistent context for AI models. While innovative, their solution lacks multi-agent orchestration and human-curated knowledge management. Conversely, this open-source project has been deployed in production settings, supporting multiple agents and integrating human input—delivering a more comprehensive and adaptable memory system.

Technical Composition

Built with a lightweight stack, the system leverages Node.js, SQLite’s FTS5 for full-text search, MCP for protocol standardization, and Express.js for server management. Notably, it foregoes more complex and costly vector databases or reliance on cloud services, making it accessible and cost-effective—approximately $60 per month for three premium AI agents with persistent memory.

Open-Source Resources

The project comprises two main repositories:

• Knowledge Base Server (The AI ‘brain’): GitHub Link
• Agent Orchestrator (Multi-agent management): GitHub Link

Community Reception and Future Potential

The system recently garnered significant interest on Reddit, demonstrating a strong demand for persistent, context-aware AI solutions. Compatibility with any AI supporting MCP, along with a custom connector for popular models like ChatGPT and Claude.ai, broadens its applicability.

Conclusion

This open-source knowledge base system offers a practical, self-sufficient approach to endow AI agents with true persistent memory. Its design emphasizes flexibility, human curation, and multi-agent coordination—making it a valuable tool for developers, researchers, and organizations seeking more capable and context-aware AI assistants. If you’re interested in implementing or customizing this solution, community support and ongoing development are readily available.

For further information or assistance in setting up your persistent AI memory system, feel free to reach out and explore the repositories linked above.