One server. One SSO. The whole enterprise AI stack — auth, LLM routing, AI apps, portal, Git, monitoring, logging, and backup. Deploy in 30 minutes.
No assembly required. No vendor lock-in. Just a complete enterprise AI platform that runs on your own server.
14 open-source components pre-integrated: auth, LLM routing, AI apps, portal, Git/CI, monitoring, logging, backup. No "glue it yourself" work.
One Keycloak account with AD/LDAP federation signs into every product automatically. Employees log in once, get every AI tool.
Fully self-hosted. Model calls, prompts, documents and user data stay inside your network. No third-party cloud dependency.
Phone numbers, IDs, emails, and bank cards are automatically redacted before any call reaches external LLMs. Presidio-powered.
Prometheus + Grafana monitoring, Langfuse LLM tracing, Loki unified logs, and IM alerting via DingTalk, WeCom, or Feishu.
Deploy, maintain, and troubleshoot the entire platform through natural language. Works with WorkBuddy, OpenClaw, or Claude Code.
14 production-grade components, orchestrated by Docker, managed through a unified admin portal.
| Layer | Component | Purpose |
|---|---|---|
| Auth | Keycloak | SSO / OIDC, AD/LDAP federation or local accounts |
| LLM Routing | NewAPI | Channels, keys, quotas, audit, cost tracking |
| PII Redaction | LiteLLM + Presidio | Auto-redact sensitive info before model calls |
| AI Apps | Dify | Visual AI app builder + RAG knowledge base |
| Portal | Ghost | Company announcements & news portal |
| Source / CI | Gitea + Runner | Internal Git hosting + Actions automation |
| Client | DeepChat | Desktop AI client (Windows / macOS / Linux) |
| Admin | AI Admin Center | Unified management dashboard + audit + reports |
| Gateway | MCP Gateway | Skill/MCP marketplace + knowledge retrieval |
| Monitoring | Prometheus + Grafana | Container resource monitoring + dashboards |
| Tracing | Langfuse | LLM call tracing, latency, tokens, cost |
| Logging | Loki + Promtail | Aggregated, searchable container logs |
| Alerting | Alertmanager | Alert routing to IM webhooks |
| Backup | Scripts + Admin UI | Daily full backup + one-click restore |
Three-tier architecture: Auth → Gateway → Services. Everything connected through a single Docker network.
Two ways to deploy. The AI Agent method is recommended — it reads the deployment guide, asks you for parameters, and configures everything automatically.
Install any AI coding agent (WorkBuddy, OpenClaw, Claude Code, etc.), point it at the repository, and let it deploy the entire platform for you.
Supports Windows (Docker Desktop), Linux (Ubuntu 24.04+), and offline deployment (pre-built Docker images included).
Free. Open source. No vendor lock-in. Your data stays on your server.