Soma

All your knowledge.
One search away.

A local knowledge-base search engine that understands your documents, images, audio and video. Nothing leaves your machine.

Download for free Learn more ›
The mental model

Three steps. That's it.

adda project
→
syncmake it ready
→
searchfind anything
Why Soma

Quietly powerful.

Hybrid search that just works.

One natural-language query expands into keywords, semantic vectors and a hypothetical document — fused and reranked by a local LLM for uncannily good results. No degraded answers.

Private by design.

Everything — indexing, OCR, embeddings, reranking — runs on your machine through a managed local runtime. Offline-friendly, always. Zero telemetry, by construction.

Sees and hears your files.

Vision extraction, OCR and transcription make screenshots, scans, audio and video recordings as searchable as plain text.

Native on every platform.

Native executables for Windows x64, Linux x64 and macOS ARM64. Instant startup, single binary — no installers, no dependencies.

Speaks your language.

Multilingual indexing that tokenizes your material, with careful CJK handling — even one- or two-character queries stay retrievable.

CLI, server, browser.

Use the terminal, run an HTTP service with a built-in web interface, or automate through one clean JSON endpoint that mirrors the CLI.

Get started

Up and running in a minute.

# Windows
winget install AwesomeDog.soma

# macOS
brew tap AwesomeDog/tap && brew trust AwesomeDog/tap && brew install AwesomeDog/tap/soma

# Linux
curl -fsSL https://github.com/AwesomeDog/soma/releases/latest/download/soma-linux-x64 -o soma && chmod +x soma && sudo mv soma /usr/local/bin/

# Then
soma project add ~/notes --name notes
soma sync
soma search "how does auth work"

Runs on 8 GB RAM · 24 GB recommended for OCR, vision & transcription · GPU optional