🔀 Chat Switchboard

Multi-Model AI Chat Platform — Self-Hosted, Extensible, Fast

A self-hosted AI chat interface that routes conversations across multiple LLM providers. Built with a Go backend for performance and a clean vanilla JS frontend inspired by Open WebUI.

---

What It Does

• Multi-provider routing — Connect OpenAI, Anthropic, Venice.ai, Ollama, or any OpenAI-compatible API. Switch models per-conversation.
• Streaming responses — SSE-based streaming with stop button, thinking block display, and full markdown rendering.
• Per-user API keys — Each user manages their own provider keys. Admins can set global keys shared across the instance.
• Admin panel — User management, global provider config, model visibility, registration toggle, usage stats.
• Self-hosted — Single Docker image, PostgreSQL backend. No external dependencies at runtime.

---

Quick Start

Docker (Recommended)

git clone https://git.gobha.me/xcaliber/chat-switchboard.git
cd chat-switchboard
cp server/.env.example server/.env
# Edit .env: set DATABASE_URL, JWT_SECRET, etc.
docker-compose up -d

Access at http://localhost:3000. First registered user becomes admin.

Manual

# Backend
cd server
cp .env.example .env
go build -o switchboard .
./switchboard

# Frontend — serve src/ with any HTTP server
cd ../src
python3 -m http.server 8080

Point the frontend at the backend by setting the API base URL (defaults to same-origin).

---

Architecture

┌──────────────────────────┐
│  Frontend (Vanilla JS)   │
│  4 files + vendor libs   │
│  No build step required  │
└───────────┬──────────────┘
            │ REST + SSE
            ▼
┌──────────────────────────┐
│  Go Backend              │
│  ├── Auth (JWT + refresh)│
│  ├── Chat CRUD           │
│  ├── Completion proxy    │
│  ├── Provider management │
│  ├── Admin endpoints     │
│  └── Static file server  │
└───────────┬──────────────┘
            │
        PostgreSQL

Frontend (src/)

File	Size	Purpose
api.js	240 lines	HTTP client, token management, auto-refresh on 401
app.js	490 lines	State, init flow, auth, chat CRUD, event wiring
ui.js	510 lines	DOM rendering, streaming, modals, formatting
debug.js	550 lines	Console/network intercept, state inspector (Ctrl+Shift+L)
vendor/	62KB	marked.js + DOMPurify (local, CDN fallback)

No framework. No build step. No node_modules. Works in disconnected environments with vendor libs baked in.

Backend (server/)

Go 1.22, ~26 source files. Key packages:

• handlers/ — Auth, chats, messages, completions, API configs, admin
• middleware/ — JWT auth, admin role check, rate limiting, error handling, logging
• providers/ — OpenAI and Anthropic adapters with streaming support
• models/ — Database models
• database/ — PostgreSQL connection + migration runner
• config/ — Environment-based configuration

API Surface

Route	Method	Auth	Description
/health	GET	—	Health check + version
/api/v1/auth/register	POST	—	Create account
/api/v1/auth/login	POST	—	Get JWT tokens
/api/v1/auth/refresh	POST	—	Rotate access token
/api/v1/chats	GET/POST	✓	List / create chats
/api/v1/chats/:id	GET/PUT/DELETE	✓	Chat CRUD
/api/v1/chats/:id/messages	GET	✓	Message history
/api/v1/chat/completions	POST	✓	Streaming SSE or sync JSON
/api/v1/models/enabled	GET	✓	Available models
/api/v1/api-configs	GET/POST/DELETE	✓	User provider management
/api/v1/profile	GET/PUT	✓	User profile
/api/v1/settings	GET/PUT	✓	User settings (persisted)
/api/v1/admin/*	various	admin	User/provider/model/settings management

---

Configuration

All via environment variables (see server/.env.example):

Variable	Default	Description
PORT	8080	Backend listen port
DATABASE_URL	—	PostgreSQL connection string
JWT_SECRET	—	Token signing key
JWT_EXPIRY	15m	Access token TTL
REFRESH_EXPIRY	7d	Refresh token TTL
CORS_ORIGINS	*	Allowed origins
REGISTRATION_ENABLED	true	Allow new signups

---

Deployment

Docker Compose (unified)

The provided Dockerfile is a 3-stage build:

1. golang:1.22-bookworm — compiles Go backend
2. node:20-alpine — downloads vendor JS libs via npm pack
3. nginx:1-alpine — serves frontend, proxies /api/ to Go backend

Vendor libs are baked into the image at build time — no CDN access needed at runtime.

Reverse Proxy

If running behind nginx/Caddy, proxy /api/ and /health to the Go backend. Serve src/ as static files.

---

Development

# Backend (hot reload with air)
cd server
go install github.com/air-verse/air@latest
air

# Frontend — just edit files, hard-refresh browser
# Debug: Ctrl+Shift+L opens debug modal

Database Migrations

Migrations in migrations/ run automatically on startup. Current schema:

• 001_full_schema.sql — users, chats, messages, api_configs
• 002_refresh_tokens.sql — token rotation
• 003_global_settings.sql — admin settings table
• 004_model_configs.sql — per-model enable/disable

---

Roadmap

See ROADMAP.md for the full plan. Next up:

1. WebSocket hub — real-time message delivery, typing indicators
2. Channels — multi-user + AI chat rooms with @mentions
3. Plugin system — Go-native extensions, installable via admin UI
4. Notes & Knowledge Base — markdown notes, document upload, RAG via pgvector

---

License

MIT — build anything, including commercial products.

---

Repository: https://git.gobha.me/xcaliber/chat-switchboard