Stash

Stash is an open-source, self-hostable platform designed to provide AI agents and humans with a persistent, shared memory space. It addresses the challenge of agents forgetting previous work by capturing every agent session—including prompts, tool calls, and artifacts—and compiling them into a unified context graph. This context graph is accessible as a virtual filesystem (VFS), enabling both agents and humans to read, search, and write data in real time, ensuring that knowledge and context are preserved and easily retrievable across sessions.

The platform connects to various external tools and data sources, including GitHub, Google Drive, Gmail, Notion, Slack, Jira, Gong, and others. Through these integrations, Stash aggregates data and agent activity into one searchable and editable space. The system supports real-time editing and synchronization, allowing agents and humans to work on the same files simultaneously. It is designed for use cases such as engineering documentation, company knowledge bases, research repositories, operational playbooks, brand guidelines, and personal knowledge management, making it suitable for teams and individual operators alike.

Stash offers three main retrieval methods for stored knowledge: a curated wiki of linked pages, vector search for semantic queries, and grep-style exact search. This ensures that information is accessible by meaning or by precise text match, and that memory is both comprehensive and easy to navigate. The platform is delivered as open-source software under the MIT license, is self-hostable to avoid vendor lock-in, and can be accessed through a CLI and MCP server. Its design emphasizes portability and ownership of data, supporting collaborative workflows between agents and humans in a single, unified environment.