I made kioku-mesh, which shares long-term memory for AI agents across multiple PCs and across multiple agents. kioku (記憶) means memory in Japanese.
I built kioku-mesh because I often work across my home PC and office PC over a VPN. A common pain for me was that I would do some development work on one machine, then continue on another machine, but the agent on the second machine had no memory of the previous context. I also found it painful that I could not easily remember when and why certain decisions were made. Another frustration was when I split work among multiple agents, for example one agent coding and another agent reviewing. The secondary agent often had to read the whole folder again from scratch, which made the workflow feel slow.
There are already tools that give long-term memory to AI agents, but many of them either require SaaS or are limited to one local machine. I wanted something that could be shared across my own PCs while still staying inside my LAN/VPN.
kioku-mesh uses Zenoh for the mesh. Zenoh + RocksDB acts as the source of truth, so when I add a new PC to the mesh, it can read memories that were saved on the other machines.
Install:
uv tool install kioku-mesh
kioku-mesh init --mode local
kioku-mesh save "Today lunch is Onigiri"
kioku-mesh search "Onigiri"
kioku-mesh mcp install --client claude-code
After mcp install, agents can call save_observation and search_memory as normal MCP tools.
Current limitations: This is still 0.x and experimental, so breaking changes may happen. Development is mainly Linux-focused, and I have not tested macOS or Windows enough yet. This is intended to be used only inside a trusted network. If you expose it to the public internet, Zenoh key-value data may be visible from outside through port 7447. I strongly recommend using it only in a closed network such as LAN/VPN/Tailscale.
Links:
- PyPI: https://pypi.org/project/kioku-mesh/
- GitHub: https://github.com/h-wata/kioku-mesh
- Demo: https://github.com/h-wata/kioku-mesh/blob/main/docs/assets/demo.gif
I would especially like feedback on what would be needed to make this useful for small teams in the future. I am also curious whether the setup and configuration feel too complicated when people actually try it.
h-wata
/
kioku-mesh
Shared memory for AI coding agents, across tools and machines. Local-first SQLite, optional Zenoh+RocksDB mesh, MCP-native.
Shared memory for AI coding agents, across tools and machines.
kioku (記憶) means memory.
kioku-mesh gives coding agents a shared memory store. Claude Code, Codex CLI Gemini CLI, and other MCP clients can save and search the same observations from one machine or from several machines on a trusted LAN/VPN mesh.
The default setup is local and needs no daemon. Mesh mode is available when you want the same memory pool replicated between hosts.
Why kioku-mesh
Coding-agent context gets fragmented across machines: which laptop did that work what did the agent on the other host decide, and why does a secondary agent have to re-read everything from scratch just to give a quick second opinion? kioku-mesh keeps that memory in one shared pool so any agent, on any of your machines, can recall it.
Unlike long-term memory tools that store everything in one place, the shared pool is a peer-to-peer…
Top comments (0)