Vane

Set up your own server · Instructions checked 2026-10-17

1. Follow the official Docker installation and keep the published Vane UI port bound to localhost. Its full image includes SearXNG; persist the data volume.
2. Prepare a local Ollama server with downloaded qwen3 and nomic-embed-text models. Configure a private container-to-runtime connection as described by the provider; do not expose an unauthenticated model port to untrusted networks.
3. Open localhost:3000, configure the Ollama provider using its reachable local base URL, and choose local chat and embedding models. Do not add paid provider keys.
4. Ask a short web question and read the linked sources. If a search engine blocks or rate-limits requests, preserve those limits; no bypass or complete search coverage is promised.

Docker/Compose and a local model runtime require sufficient CPU/RAM or GPU memory and disk space for chat and embedding models. The provider gives no universal minimum; model/context size drives the main cost. This local model route has no required paid API or software subscription. You provide hardware, storage, electricity and download/network access. Self-hosted web search still sends queries to search engines and websites; it is not offline. Paid model and hosted search providers are separate.