Inhouse AI: a private, self-hosted AI assistant (open source)
Every time someone pastes a contract or client spreadsheet into ChatGPT, that data leaves your network. A client came to me with exactly this problem: they wanted ChatGPT-style help, but their documents couldn't leave their own systems.
So I built Inhouse AI. It runs entirely on your own hardware:
- Local models, so nothing is sent to an AI provider
- Q&A over your own documents, with source citations
- Multi-user, with private knowledge bases
- Secure remote access with zero open ports
- One Docker Compose file, from a laptop to a GPU server
Built on Open WebUI, LiteLLM, Ollama and Postgres with pgvector, and benchmarked on an ordinary laptop GPU.
MIT licensed and free to use: https://github.com/razajahangir/inhouse-ai
Need it set up for your team or adapted to your tools? Feel free to reach out.
GitHubGitHub - razajahangir/inhouse-ai: self-hosted llm rag local-llm private-ai open-webui litellm ollama pgvector docker-compose cloudflare-tunnel retrieval-augmented-generationself-hosted llm rag local-llm private-ai open-webui litellm ollama pgvector docker-compose cloudflare-tunnel retrieval-augmented-generation - razajahangir/inhouse-ai
commit 0decb79