Dossier 05 · Robotics, the frontier
On-board memory for robots
A robot cannot hand its memory to the cloud. Latency breaks its reflexes, and a network drop must not make it forget. What it learns about your factory, your home or the street must stay on hardware you control, auditable by a human. Robotics is young: this is the time to design its memory.
Mnemosyne OS provides this memory: encrypted vaults on the machine itself, semantic search with no network, and a human who decides what is kept or forgotten.
Robotic memory : on-board memory for robots and autonomous systems: local, encrypted, no network needed, governed by a human. Mnemosyne OS is a memory engine built to live where the robot lives.
We are looking for robotics teams, industrial or academic, to explore it with us.
Windows, macOS and Linux.
Our thinking Our approach: a memory that follows the pace of the motors
A design direction. The pieces described here already run in Mnemosyne OS on a computer. Here is how the same architecture would work on a machine that moves.
Motors running
During work, memory stays out of the control loop. No consolidation, no re-indexing: the compute serves perception, navigation and reflexes. Memory answers fast, from the vault linked to the task. The rest of the database sleeps.
Motors at rest, charging
This is the time for the dream engine, the consolidation Mnemosyne OS already runs during downtime. It takes the day’s raw data, removes the noise, recomputes the embeddings and creates new links. It is the machine version of consolidation during sleep, done when the compute is free.
Spines, not one big brain
Memory is split into isolated, encrypted vaults: navigation on one side, user profiles on the other, by domain and by level of criticality. A search only wakes the relevant vault. These vaults and these spines already exist in the OS. What remains is placing them on a robot.
Mnemosyne OS runs on computers today. This is the plan we would follow with a robotics partner.
Questions
What a robotics team asks before anything else
- What hardware does it run on today?
- On a computer: Windows, macOS and Linux, on x86-64. The ARM64 port, for an embedded board such as a Jetson, is one of the first things a partnership defines.
- Can our stack query the memory?
- Yes. The application opens three entry points on the machine: a REST gateway for status, ingest and queries, and two WebSockets, one for the IDE extension, one for the SDK and cartridges. All three listen locally, never on the network. An MCP server is also published on npm, under the MIT licence.
- How does a collaboration start?
- With a conversation about your machine and its constraints: what it must keep, what it must forget, what has to survive a dropped link. You have the robot. We bring a memory engine that already runs, and a core you can reach without an interface.