Zilliz, a company that has developed an open-source vector database called Milvus, has now announced the availability of Memsearch, a lightweight library that will provide AI agents with permanent and long-term memory. This memory will be fully transparent and readable by humans. Memsearch is a tool that will address one of the most important problems facing the developing AI agent ecosystem. This problem is the lack of permanent memory for AI agents. Memsearch is now available as an open-source tool. It is available for use by developers who want to create AI-based applications or autonomous AI agents. Memsearch is based on a memory system that was first developed for OpenClaw, an autonomous AI agent that quickly became very popular on GitHub. However, the memory system developed for OpenClaw has now been refined and separated from OpenClaw. It is now available as Memsearch, which will function with any AI agent.
Addressing the Memory Gap in AI Agents
In current AI agents, continuity is a problem. After the end of each session, the AI agent forgets its context, preferences, and past decisions, prompting the user to continuously provide the information. Although some solutions have been developed to address this problem, most solutions depend on proprietary storage mechanisms, which cannot be viewed, edited, or customized by developers.
In Memsearch, the problem is addressed differently, focusing on transparency and developer control. Unlike most solutions, which store AI agent memory using proprietary mechanisms, Memsearch stores memory as plain text files, which can be viewed, edited, and controlled by developers. The files are then indexed using the Milvus database, facilitating semantic search and retrieval.
This design ensures that developers maintain full visibility into what an AI agent remembers, making it easier to debug incorrect memories, refine stored knowledge, or migrate data across platforms.
Also Read: CrowdStrike and Perplexity Extend Enterprise-Grade Security to Comet Enterprise AI Browser
Building Transparent and Portable AI Memory
Another important aspect of the Memsearch architecture is the emphasis that has been placed on openness and portability. For instance, the use of text files in the memory layer enables developers to use software development techniques, which can be used to manage the memories that are being developed. Furthermore, developers can update the memories that are being stored without having to train the models, which can be used to update AI knowledge.
“Memory is the missing layer in the AI agent stack. Developers deserve to know what their agents remember, fix it when it’s wrong, and carry it forward without lock-in. memsearch is our answer to that—transparent, portable, and built on the open-source foundation the community already trusts.” – Jiang Chen, Head of Developer Relations, Zilliz
Key Capabilities Designed for Developers
The release of Memsearch introduces several features aimed at improving the reliability and usability of AI agents:
- Human-readable memory: All stored knowledge is maintained as simple text files that developers can easily inspect.
- Editable and correctable records: Incorrect or outdated information can be fixed by editing the file directly, with Memsearch automatically detecting and re-indexing updates.
- Collaborative workflows: Because the memory layer exists as standard files, teams can manage it using tools such as Git for version control and peer review.
- Portability across platforms: Moving an AI system between machines, frameworks, or cloud environments requires only copying the memory files.
- Flexible integration: The library is designed to work with a wide range of AI agent frameworks with minimal configuration.
These capabilities help developers maintain greater control over AI behavior while also reducing reliance on proprietary data formats.
Plugin Support for AI-Assisted Development
Along with the Memsearch plugin, Zilliz also announced the memsearch ccplugin, which has been specifically developed for Claude Code, an AI coding assistant developed by Anthropic. The plugin would allow coding sessions to retain memory, saving key information like architectural decisions, debugging, and team preferences.
The plugin would automatically fetch historical context and inject it into the prompt of the assistant, helping developers pick up on previous conversations and decisions made, thereby helping them be more productive with AI coding assistants.
