Your files.
Your answers.
Your machine.
File Palaces mines your documents into a local vector palace and lets you chat with them via streaming AI. No cloud. No subscription. Your data stays exactly where it belongs.
Searching files is not understanding them.
You have the documents. The answers are in there. But grep, find, and Ctrl+F were built for a different era. File Palaces answers your questions and shows you exactly where the answer came from.
$grep -r "renewal clause" ./contracts/
Binary file contracts/Q3_signed.pdf matches
Binary file contracts/vendor_2024.docx matches
$ # now what?
// open each file, Ctrl+F, repeat...
Three steps. One action each.
Drop a folder
Drag any folder onto the app or use the picker. File Palaces creates a Wing — a named region of your palace dedicated to that folder tree.
Background indexing
The sidecar extracts text from every file, chunks it, embeds it with a local model, and stores it in ChromaDB. A watchdog keeps the index current.
Streaming answers
Ask questions in plain language. The LLM retrieves the most relevant chunks, synthesises an answer, and shows you exactly which files it cited.
Everything you need.
Nothing you don't.
Local‑first
Everything runs on your machine. Data never leaves. No cloud dependency, no telemetry, no subscriptions.
Multi‑LLM
Ollama (offline), OpenAI, Anthropic, and any OpenAI‑compatible endpoint. Swap providers in seconds.
Hybrid search
BM25 keyword + ChromaDB dense vectors fused with Reciprocal Rank Fusion. Better recall than either alone.
Live sync
A watchdog observer re-mines changed files automatically. Your palace stays current as your folders evolve.
Palace Map
Visual overview of all Wings, Rooms, file‑type breakdowns, and drawer counts. Know what's indexed.
Source citations
Every answer links back to the exact chunk it came from. Click to preview the surrounding context.
Index anything.
Documents, spreadsheets, audio, video, email, archives, and live web pages. If it has text in it, File Palaces can mine it.
OCR FOR SCANNED PDFS · COMING SOON
Your LLM, your choice.
File Palaces routes through LiteLLM and supports every major provider. Switch at any time from Settings.
Fully local inference. Pull any model with one command. Works entirely offline.
GPT‑4o, o3, and other models. Set your key in Settings → LLM.
Claude Opus, Sonnet, Haiku. Plug in your Anthropic key and go.
Any OpenAI‑compatible endpoint. Point to LM Studio, Together, Groq, or your own proxy.
Nothing leaves your machine.
Extraction, chunking, and embedding all run locally. Your palace lives in your app data directory. No analytics, no telemetry, no cloud sync. Ever.
Ready to build your palace?
Read the docs and get started in minutes.