RememberOS
The shared memory operating system for AI agents and humans — a common layer where agents and people store, search, govern, and share knowledge together. Most memory systems ask "how can an agent remember?" RememberOS asks "how can agents and people remember together?"
Most of this is available today: persistent memory, semantic search, collections, agent identities, provenance, and shared collections with access control — see Multi-agent collaboration to use them. Still ahead: distinct human vs. agent principals, cross-workspace sharing, and deeper intelligent-memory operations.
The problem: memory silos#
Today's agents are trapped in silos. Each keeps its own context window, vector store, or knowledge base. That helps one agent remember across its own conversations, but an insight found by one agent stays invisible to another — and to the people working alongside them. Knowledge gets fragmented, duplicated, and isolated. As you add agents, the memory problem becomes a collaboration problem.
The idea: one shared memory layer#
Instead of every agent maintaining isolated memory, agents and humans connect to a common memory layer where knowledge is stored, searched, organized, governed, and shared. The memory outlives any individual agent, model, framework, or conversation.
Agents come and go. People come and go. Memory remains.
An OS for knowledge, not just a database#
RememberOS provides knowledge as first-class infrastructure:
- Persistent memory & semantic search
- Shared collections — organize knowledge with clear ownership
- Access control & governance — decide who can read, write, and curate
- Principal identities — agents and humans are first-class participants
- Provenance — every memory records who (which agent or person) contributed it
- Intelligent memory — consolidation, summaries, contradiction surfacing (over time)
Collections: ownership and access#
Knowledge lives in collections, and access is granted to principals — agents or people:
| Collection | Accessible by |
|---|---|
| Customer Support | Support Agent · Email Agent · Billing Agent · the support team |
| Product Knowledge | Research Agent · Docs Agent · Dev Agent · the PM |
| Personal Memory | Personal Assistant · Voice Agent · just you |
Collaboration — agents and humans#
The goal isn't only retrieval; it's collaboration. On a common substrate, principals learn from each other and build on existing knowledge: an agent surfaces what a teammate captured, a person curates or corrects what an agent wrote, and discoveries by one agent become available to the next. Knowledge becomes a shared organizational asset rather than a private implementation detail.
How your data maps#
Nothing breaks. Today's single-key usage becomes a one-principal organization with your collections private by default — identical behavior — and from there you can mint additional agent identities and share collections with other agents and people. See the isolation model and keys for how identity and access work.