Skip to content
WindsanderPublic

About

🌌 Mebular 忆云 · 智能体共享记忆与任务协作 | One shared memory and task graph for all your AI agents and devices: automatic or on-demand, secure, controllable, and decentralized!

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

162 stars

Watchers

13 watching

Forks

Latest commit

 

History

1,168 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mebular

One memory. Every assistant. Every device.

A distributed memory network for your AI assistants.
Memory and tasks stay on your devices; devices sync directly.

CI License: MIT Node.js >=20 MCP server DSH plugin

Start in 5 minutes · How it works · Website · 中文

No hosted account · No central coordinator


Star map baseline: the device network before hovering Task lanes and a fleet event: blue can send, gold can receive, gray unreachable

The map at rest: devices and their shared-subscription edges. · On hover: task-domain lanes plus live fleet traffic.
Blue and gold show this device's two-way task-domain reachability, not peer-to-peer links.

One memory

Why Mebular: one shared graph instead of a silo; local-first, direct sync, auditable, MCP

Memory Runtime: lightweight MCP and immersive access share one graph

0.3 adds an optional runtime above MCP. Default off; local Qwen optional.

Memory Runtime guide · DSH · Hermes · OpenChamber

Get started

DSH plugin integration: the DSH sidebar loads the same Mebular Console, and the official MCP client reaches the same daemon

Desktop app setup: 1 Start deployment, 2 Create or join a Cloud, 3 Invite the next device

Other ways to start: Browser, CLI and Agent, all sharing one graph

Desktop app · v0.3.2 Download & first launch
Unsigned release
macOS Universal · Windows x64 · Linux x64
Download from GitHub Releases
Verify with SHA256SUMS.txt. macOS: right-click → Open. Windows: More info → Run anyway. Bundles Electron and its runtime.

Or install the CLI and server:

npm install -g @mebular/core
mebular serve                                  # browser console
fleet quickstart --daemon                      # CLI start
mebular print-config --client opencode         # MCP config

How it works

Mebular network overview: each machine runs one node; nodes sync memory and tasks directly; add devices with a QR code or token

Write with memory_write, hand off with task_submit — the same graph, from any assistant.

Pairing flow: device A invites, device B joins with the QR code or token, then connection and the default grant happen automatically

Docs

Start Learn Run Trust
Quickstart
Install & pair
Architecture
Nodes & sync
Console
Local GUI
Limitations
Sensitive data
Agent skills
Setup playbooks
Fleet ops
Multi-machine
Release SOP
Release steps
Sealing docs
Protocol rules
Changelog · Roadmap
Released & planned
Contributing
Quality gates
Security
Report a vuln
Website
Product site

Use with DSH
Install the plugin from npm into DSH's web profile:
dsh plugin --profile web add @mebular/dsh-plugin@0.3.2
The App and the DSH Plugin are two equal entries to one local Mebular runtime: the plugin defaults to attach and connects to an existing App/CLI/service without starting or downloading anything. If no runtime exists, the panel explicitly offers a pinned shared-runtime install. The sidebar panel embeds the same Console through ?embed=dsh (loopback only), and the official MCP client reaches the same graph. The plugin is an independent adapter with its code isolated in packages/dsh-plugin; it is published by its own npm job, while its version follows the unified release train. For plugin development, install from packages/dsh-plugin instead.
Listed on awesome-dsh-plugin under memory.
DSH plugin guide → · Integration notes →

Who does what
Who Handles
You One-time pairing, Cloud choice, invites, subscriptions.
Your assistant Memory reads and writes, task claim and completion.
Mebular Discovery, direct / relay / hole-punch, retries, certificates, service autostart.
Use it from code
import { Mebular } from 'mebular';

const mebular = new Mebular({
  storagePath: './store.jsonl',
  deviceId: 'device-a',
  network: { enabled: false },
});

await mebular.initialize();

Runnable examples: examples/quickstart writes and recalls a preference; examples/two-devices demonstrates offline sync.

v0.3.2 · early · MIT · © 2026 Sobert Li

About

🌌 Mebular 忆云 · 智能体共享记忆与任务协作 | One shared memory and task graph for all your AI agents and devices: automatic or on-demand, secure, controllable, and decentralized!

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

162 stars

Watchers

13 watching

Forks

Releases

Sponsor this project

Packages

Used by

Contributors

Languages