GenMult

GenMult is a self-hosted multi-agent workspace for building, running, and coordinating AI assistants across web, desktop, and external chat channels.

It started from the idea of giving every agent its own long-running context, tools, files, memory, and delivery channels. The current project has grown into a full-stack system with a Go backend, a Vue web client, an Electron desktop app, containerized agent workspaces, AgentHub collaboration rooms, and channel adapters such as WhatsApp.

What GenMult Does

• Run multiple AI agents with separate identities, models, settings, memory, and permissions.
• Chat with agents from the web UI, desktop shell, local sessions, and connected channels.
• Use AgentHub rooms to coordinate humans and agents in the same workspace.
• Give agents isolated containers for file operations, command execution, MCP tools, and longer-running work.
• Keep conversation history, memory, scheduled tasks, and channel routes in a database.
• Configure providers, models, channels, storage, containers, and tools from a graphical interface.
• Connect external channels including Telegram, Discord, Feishu/Lark, DingTalk, WeChat-family adapters, Matrix, Email, and WhatsApp.

Main Components

Component	Stack	Default Port	Purpose
Server	Go, Echo, Twilight AI SDK	8080	REST API, auth, database, agents, channels, containers
Web	Vue 3, Vite, Pinia, Tailwind CSS	8082	Browser-based control panel and chat workspace
Desktop	Electron, electron-vite	local app	Desktop shell that reuses the web client and local server
Database	PostgreSQL or SQLite	varies	Users, bots, rooms, messages, memory, providers, settings
Vector / Search	Qdrant, sparse search	optional	Long-term memory retrieval and hybrid search
Containers	Docker, containerd, Kubernetes, Apple Virtualization	varies	Isolated workspaces for agent tools and files

Features

Agent Workspace

• Multi-agent and multi-user chat.
• Per-agent model, provider, memory, channel, and permission settings.
• AgentHub rooms for shared collaboration, mentions, room messages, and agent membership.
• Long-term memory providers with history compaction and retrieval hooks.
• Scheduled tasks, heartbeat sessions, discuss mode, and subagent workflows.

Channels

• Built-in web and local chat.
• WhatsApp support with QR login, media handling, group routing, and bot mention behavior.
• Channel adapters for Telegram, Discord, Feishu/Lark, DingTalk, QQ, Matrix, Misskey, WeCom, WeChat, WeChat Official Account, and Email.
• Source-aware access control for deciding who can trigger which agent.

Tools And Runtime

• MCP support for external tool servers.
• Container-backed command execution and file editing.
• File manager, attachments, media assets, and browser-facing workspace views.
• Provider configuration for OpenAI-compatible APIs, Anthropic, Google, GitHub Copilot, Codex-style clients, TTS, and search providers.

Apps

• Web dashboard for agent management, chat, settings, channels, and AgentHub.
• Electron desktop app for a local client experience.
• Docker Compose deployment for server, web, database, migrations, and optional memory services.

Quick Start

Requirements:

• Docker and Docker Compose
• Git

Clone the repository:

git clone https://github.com/ZihengXiong/GenMult.git
cd GenMult

Create a local config file:

cp conf/app.docker.toml config.toml

Edit config.toml for your admin account, database, model providers, container runtime, and channel settings.

Start the stack:

docker compose up -d

Open the web client:

http://localhost:8082

The backend API listens on:

http://localhost:8080

Desktop Development

Install frontend dependencies:

corepack enable
pnpm install

Run the desktop app in development mode:

pnpm --filter @memohai/desktop dev

Build desktop packages:

pnpm --filter @memohai/desktop build

The desktop package still reuses the web renderer internally, so web UI changes are shared by both browser and desktop builds.

Web Development

Run the web client:

pnpm --filter @memohai/web dev

Build the web client:

pnpm --filter @memohai/web build

Backend Development

Run the Go server locally:

go run ./cmd/agent

Regenerate database code after SQL changes:

mise run sqlc-generate

Generate OpenAPI and TypeScript SDK artifacts after API changes:

mise run swagger-generate
mise run sdk-generate

Repository Layout

cmd/                 Go entry points for server, bridge, MCP, and CLI
internal/            Backend domain modules
apps/web/            Vue web client
apps/desktop/        Electron desktop shell
packages/            Shared UI, SDK, icons, and config packages
db/                  PostgreSQL and SQLite migrations and queries
conf/                Example configuration and provider templates
docker/              Production Dockerfiles and nginx config
docs/                Documentation site
deploy/              Kubernetes deployment examples

Security Notes

• Do not commit .env, config.toml, keys, local database files, generated backups, or deployment secrets.
• Channel credentials and model provider API keys should stay in local config, environment variables, or a secret manager.
• Each agent workspace should be treated as an execution boundary. Review container privileges before exposing a deployment publicly.
• For production deployments, change the default admin password, enable HTTPS at the reverse proxy, and restrict inbound ports.

Current Status

GenMult is under active development. Some internal package names and historical paths may still use the original module naming while the product identity moves toward GenMult.

License

See LICENSE.