Run your own instance
Insight AI can run entirely on your own hardware. A self-hosted instance gives you the same dashboard, ingestion pipelines, and chat experience as the hosted product — with all data, documents, and AI inference staying inside your own environment. Nothing leaves your network: the LLM runs locally via Ollama, and documents are indexed into a local vector database.
What’s in the box
Section titled “What’s in the box”docker compose up starts the whole platform as containers:
| Service | Role |
|---|---|
| Backend (FastAPI) | The REST API the dashboard and your scripts talk to — the only service exposed to the host |
| PostgreSQL | Application database (users, chats, sources, pipelines) |
| Weaviate + text2vec transformer | Vector database and embedding model powering semantic search |
| Ollama | Runs the AI models locally |
| Redis + Celery worker | Background job queue that executes ingestion runs |
| Apache Tika | Extracts text from PDFs, Office documents, and more |
| Samba (optional) | A bundled SMB file server for testing data sources without real infrastructure |
The React dashboard (frontend) is run separately during evaluation — see installation.
Requirements
Section titled “Requirements”- Docker Engine with the Compose plugin, running on Linux (recommended) or macOS/WSL2.
- CPU & memory: 4+ cores and 16 GB RAM comfortably run the stack with a small model; more helps, especially while ingesting.
- Disk: ~10 GB for images and databases, plus room for the AI models you pull (1–40+ GB each depending on size).
- GPU (optional but recommended): an NVIDIA GPU with the
NVIDIA Container Toolkit
makes model responses dramatically faster. Without it, models run on CPU —
fine for evaluation with small models like
tinyllama. - Node.js 20+ to run the dashboard frontend.
Where to get it
Section titled “Where to get it”The source lives at github.com/insight-source/insight-engine.
Ready? Continue to installation.