LocalChat
A self-hosted RAG platform: chat with your own documents, without any of it leaving the building.
What it is
LocalChat is an open-source application for asking questions of your own documents — PDF, Word, PowerPoint, Excel, text, Markdown and images — and getting an answer from a language model that runs on your own hardware. Documents are chunked, embedded and stored in PostgreSQL with pgvector; the answer comes from a combination of semantic and lexical search, with a cross-encoder that reranks the results before the model sees them.
Why it exists
Most "chat with your documents" products send those documents to someone else's cloud service. LocalChat does the opposite: the model (via Ollama), the database and the search layer all run on infrastructure you control yourself. Nothing leaves the machine unless you deliberately turn on web search or a cloud fallback. That makes it usable in situations where it has to be that way — sensitive documents, internal knowledge, a team that doesn't want to depend on an outside party for something as basic as searching its own files.
Scope: deliberately limited
LocalChat isn't a SaaS product in the making. It's built as an appliance for a small team — up to a few dozen users — on a single node. Running multiple replicas right now deliberately breaks cache consistency and rate limiting; that's an explicit, documented choice, not a shortcoming still waiting to be fixed. Within that boundary, the path to production-grade has been taken seriously: a hardening track with fail-closed boot, authorisation enabled by default in CI, a security core tested with mutation testing, and a tested restore procedure.