AnythingLLM

One line: AnythingLLM (68k+ stars on GitHub) is a desktop knowledge-base AI: drag PDFs and notes into a workspace — it chunks, embeds and indexes them, then answers questions with the source passages cited. Built-in local model engine out of the box, cloud keys optional.

What you get

• A private knowledge base: documents → vector store → answers with citations (RAG), all on your machine
• Built-in local LLM engine (Lemonade + llama.cpp; GPU support files are fetched during install),

usable without buying any API key

• Cloud models too: OpenAI / Anthropic / OpenRouter / Google / local Ollama, etc.

(full list in the official docs)

• Isolated workspaces, Agent skills, audio transcription, scheduled tasks and more

Before you start (2 minutes)

• Windows 10+ (x64 or ARM64) or macOS (separate builds for Apple Silicon / Intel)
• Keep 5 GB+ free (app + built-in engine + GPU support files; optional meeting-transcription

model is ~2 GB more)

• Network needed while installing to fetch the engine/GPU files (vendor CDN); after that you can

run local models fully offline

• ⚠️ Official requirement: install "current user only" — "all users" is not supported; no

administrator rights needed at any point

• Not needed: Docker, Python, Node

Install

Route A: install with AiBOX (recommended)

1. Open AiBOX → app directory → AnythingLLM → Install
2. AiBOX downloads the official installer (sha256-verified, accelerated automatically in CN):

Windows uses the vendor's silent mode /S into %LOCALAPPDATA%\Programs\AnythingLLM (current user, zero elevation); macOS picks Silicon/Intel dmg by machine and copies into /Applications

Route B: manual install (official channels)

• Windows: download the installer for your architecture from

anythingllm.com/download and double-click it (keep the "current user" option); silent install (official docs, v1.16.1+):

.\AnythingLLMDesktop.exe /S                     # silent, skips the transcription model
.\AnythingLLMDesktop.exe /S /DOWNLOAD_MODEL=yes # silent + download the ~2GB transcription model

• macOS: download AnythingLLMDesktop-Silicon.dmg (M-series) or AnythingLLMDesktop.dmg

(Intel), drag AnythingLLM into Applications; or brew install --cask anythingllm

Configuration (connect a model)

1. First launch runs onboarding: pick a Workspace and name it
2. Gear icon (top right) → AI Providers:

• Built-in engine (default): works out of the box — zero configuration
• Cloud keys: paste an OpenAI / Anthropic / OpenRouter key (from that vendor's console)
• Ollama: auto-detected if Ollama is running — choose a model

1. Add knowledge: workspace → Upload Documents → drop PDF/Word/Markdown → click Ingest
2. Chat with that workspace selected — answers will cite your documents

Verification: what "success" looks like

1. The app opens and shows its version (top right; "Open storage" jumps to the data folder)
2. Create a workspace, upload a PDF, ingestion finishes with a green check
3. Ask a question only that PDF can answer → correct reply with the cited source = RAG chain works
4. Settings → AI Providers shows the model connected and replying = key configured correctly

FAQ

• "AnythingLLM cannot be closed" during install: an old instance is still in the tray —

right-click the tray icon → Quit, then retry (official troubleshooting doc)

• A previous "all users" install exists: uninstall it first with

Uninstall AnythingLLM.exe from its install folder, then reinstall for the current user

• Local model slow / no GPU: if the GPU support files failed to download during install the

app falls back to CPU — check the network and reinstall, or read *_install.log in the install folder

• Upgrading: run the new installer over the old version — data stays put; the in-app update

banner also notifies you

• Where is my data? Windows %APPDATA%\anythingllm-desktop\storage,

macOS ~/Library/Application Support/anythingllm-desktop/storage

Uninstall

• Windows (official silent flags): run

"%LOCALAPPDATA%\Programs\AnythingLLM\Uninstall AnythingLLM.exe" /S (keeps data); add /S /DELETE_DATA=yes to purge data too (irreversible)

• macOS: quit the app and drag AnythingLLM.app to the Trash,

or brew uninstall --cask anythingllm (add --zap to remove data)

• ⚠️ Workspaces, parsed documents, vectors, chat history and local models all live in the data

folder — back up before deleting

More: install docs · storage locations · uninstall