DashHub

Self-hosted AI workspace for teams. Your data, your models, your rules.

Quick Start · Features · Architecture · Contributing

---

DashHub is an open-source AI workspace that lets your entire team share access to GPT-4, Claude, Gemini, and local models through one private, self-hosted interface — without paying per seat or sending data to multiple vendors.

---

Quick Start

Prerequisites: Docker + Docker Compose v2

# Clone and start everything in one shot
git clone --depth=1 https://github.com/DashHub-ai/DashHub && cd DashHub
docker compose -f docker-compose.quickstart.yml up -d --build

Or with the installer script:

curl -sSL https://raw.githubusercontent.com/DashHub-ai/DashHub/main/quick-start.sh | bash

Once running (first build ~3 min):

Service	URL
Chat app	http://localhost:5173
Admin panel	http://localhost:5174
API	http://localhost:3000

Default credentials: root@dashhub.ai / dashhub123

Next step: Open the admin panel, create an organization, then add an AI model (OpenAI key, Ollama, or any OpenAI-compatible API).

---

Features

🤖 Multi-model, one interface

Connect OpenAI, Anthropic Claude, Google Gemini, DeepSeek, and any OpenAI-compatible endpoint — including local models via Ollama for zero-cost, fully private inference.

🔧 MCP tool servers

Plug any Model Context Protocol server into your workspace. Your AI gets access to databases, APIs, file systems, and custom tools — all through a single config.

📂 Projects + knowledge bases

Organize work into projects with shared file uploads and vector-search knowledge. Every team member working in the same project has the same context.

🤖 Custom agents

Build agents with custom system prompts, a pinned AI model, and scoped knowledge. Share them across the organization or keep them private.

⌨️ Command palette

Press Cmd+K (or Ctrl+K) to instantly jump anywhere — chats, projects, agents — or search across all your content without leaving the keyboard.

📌 Pins

Bookmark important AI responses and share them with your team. Pins persist across model switches and chat sessions.

🔒 Role-based access

Three roles out of the box: Admin (user management), Tech (model & integration config), User (chat, projects, agents). Per-project permissions on top.

🔍 Full-text + vector search

Elasticsearch powers both keyword search and semantic (embedding-based) retrieval across all chats, projects, and knowledge bases.

🗂️ S3-compatible file storage

MinIO ships in the default stack. Drop in AWS S3, Cloudflare R2, or any S3-compatible bucket by changing three env vars.

---

Architecture

┌──────────────────────────────────────────────┐
│  Browser                                      │
│  ┌─────────────────┐  ┌──────────────────┐   │
│  │  Chat app :5173  │  │  Admin UI :5174  │   │
│  └────────┬─────────┘  └────────┬─────────┘   │
└───────────┼──────────────────────┼─────────────┘
            │                      │
            ▼                      ▼
    ┌───────────────────────────────────┐
    │  Hono API server  :3000           │
    │  (TypeScript, tsyringe DI)        │
    └────┬──────────┬──────────┬────────┘
         │          │          │
         ▼          ▼          ▼
    Postgres    Elasticsearch  MinIO
    + pgvector  (search)       (files)

Stack: TypeScript monorepo · Hono (backend) · React + Vite (frontend) · Kysely ORM · fp-ts · Zod · Tailwind CSS

---

Configuration

All configuration is passed as environment variables. Copy the quickstart defaults or see apps/backend/.env.example.

Variable	Description	Default
USER_ROOT_EMAIL	Root admin email	root@dashhub.ai
USER_ROOT_PASSWORD	Root admin password	dashhub123
JWT_SECRET	Auth token signing key — change in production	—
APP_ENDUSER_DOMAIN	Domain for user access (used in emails)	localhost
DATABASE_*	PostgreSQL connection	see compose file
ELASTICSEARCH_*	Elasticsearch connection	see compose file
MINIO_*	S3-compatible storage	see compose file

---

Development setup

git clone https://github.com/DashHub-ai/DashHub && cd DashHub
docker compose up --build   # starts all services in dev/watch mode

The dev setup hot-reloads the backend on TypeScript changes and runs Vite dev servers for both frontends.

# Run DB migrations manually
cd apps/backend && npm run db:migrate

# Re-index all content in Elasticsearch
cd apps/backend && npm run es:reindex:all

---

Contributing

Issues and PRs are welcome. A few things that would make a big difference:

• Frontend for eval runner — backend + SDK are merged, needs a UI
• MCP server directory — curated list of useful MCP servers for teams
• Docker image publishing — GitHub Actions workflow to push to ghcr.io
• SSO / SAML — enterprise identity provider support
• More AI providers — Cohere, Mistral API, Bedrock

See open issues and CONTRIBUTING.md to get started.

---

User roles

Role	Can do
Admin	Manage users, assign roles, view audit logs
Tech	Add AI models, configure MCP servers, manage S3 buckets, build agents
User	Chat, create projects, use agents, upload files, pin responses

---

Roadmap

• Multi-model support (OpenAI, Gemini, DeepSeek, Ollama)
• MCP tool server integration
• Command palette (Cmd+K)
• Projects + vector knowledge bases
• External API integrations
• Eval / benchmark runner
• SSO / SAML
• Published Docker images (no-clone install)
• Audit log UI
• Plugin marketplace

---

License

MIT © DashHub.ai