Prism
A privacy-first AI command palette that searches eight self-hosted sources in parallel from one hotkey. Follow-up questions are answered from the results on screen and cited, with a local model doing the reading. Built as a native desktop app with Tauri 2 and Rust.
A Desktop App That Thinks Locally
The Vision
I wanted a single keyboard shortcut that could search everything I self-host: Obsidian notes, Paperless-ngx scans, the BookStack wiki, Nextcloud, mail. Existing tools either required cloud processing or searched one source at a time.
Why Tauri + Rust
Electron was the obvious choice, but I wanted native performance and a small footprint. Tauri 2 compiles to a native binary with a Rust backend, so the app launches instantly, uses minimal RAM, and can make system-level calls that Electron can't. The tradeoff was a steeper learning curve with Rust's ownership model.
The Connector System
Each data source is a pluggable connector with a standardized interface: search, fetch, and index. Connectors run in parallel so a single query fans out to all eight sources at once. Results are ranked on text similarity, source reliability, title match, recency, and phrase bonus, then fused across connectors by rank rather than by scores that aren't comparable.
Local-First AI
Prism connects to LM Studio or Ollama for fully offline AI. RAG queries embed the user's question, retrieve relevant chunks from Qdrant, and pass them as context to the local model. Every answer includes citations linking back to the source document. Cloud AI is available as opt-in with bring-your-own-key.
Eight Sources, One Search
A ninth connector, LM Studio, runs the model rather than answering a query. Every one of these is a service you already run, with your own credentials over your own network. Prism searches your infrastructure and supplies none of it.
Built with
See the Full Product
Visit the marketing site for features, pricing, downloads, and a free trial.