AnythingLLM Is the Self-Hosted AI Platform Non-Coders Have Actually Been Waiting For

If you're a lawyer who wants to query your entire case archive in plain English, a marketer who needs a private AI trained on internal brand docs, or an ops lead building lightweight automations without touching code — AnythingLLM is probably the first self-hosted AI platform you should look at. Not because it's the most powerful, but because it's the one where the gap between "I installed it" and "I'm actually using it" is the shortest.

Five years of active development means the rough edges a typical open-source project has at launch have mostly been filed down. The project ships a desktop app, a Docker image (a self-contained package you run on your own server or computer), and a cloud option — all running the same codebase.

At a glance

What AnythingLLM actually is

AnythingLLM is a self-hosted AI workspace. The core idea is that you bring documents, configure an AI provider, and get a private ChatGPT-like interface that can actually read your files — not just pretend to. The technical term is RAG (retrieval-augmented generation): your documents get indexed into a local database on your own machine, and when you ask a question, the system pulls the relevant chunks and sends them alongside your query to the AI model. You see source citations in the response. The model answers from your content, not from its general training.

That's the foundation. What AnythingLLM adds on top is a full platform: multiple workspaces with different document sets and different AI configurations, a user management system with role-based access, a no-code agent workflow builder, scheduled jobs, an embeddable chat widget for websites, and MCP (Model Context Protocol) support so agents can reach outside the app to external services. One self-contained Docker package covers all of it.

The AI model provider list is unusually wide: OpenAI, Anthropic (Claude), Google Gemini, AWS Bedrock, Azure OpenAI, Ollama, LM Studio, Groq, Mistral, Cohere, Together AI, and 25+ others. The local model path — Ollama running Llama 3.3 or Qwen on your own hardware — means you can run the entire stack without a single API key or a single byte leaving your machine.

What ships in the repo

Workspaces and document intelligence

• Drag-and-drop document upload: PDF, DOCX, TXT, spreadsheets, and more
• Per-workspace document intelligence (RAG) with source citations in every response
• Multiple document storage backends (the index where your files are kept for fast search): LanceDB (default, zero-config), PGVector, Pinecone, Chroma, Weaviate, Qdrant, Milvus
• Full-text and semantic search across documents

Agents and the no-code builder

• Typing @agent at the start of a message activates the AI agent in any chat
• Built-in agent skills: web browsing, web scraping, file read/write, chart generation, SQL queries, Gmail, Outlook, Google Calendar, Outlook Calendar, scheduled job creation
• Agent Flows: a visual, block-based workflow builder — no code, no config files, just drag blocks for Web Scraper, API Call, AI Instruction, Read File, Write File, and wire them together. Built-in debugger shows exactly where a flow breaks.
• MCP compatibility for extending agents to any MCP server (GitHub, Notion, Slack, and hundreds of others)
• Custom agent skills via plugin.json + handler.js for developers who want to go further

Platform and multi-user

• Multi-user with admin / manager / default role tiers (Docker mode)
• Per-user and per-workspace access controls
• Embeddable chat widget for external websites
• Full web API for custom integrations
• Desktop Pro features: Magic Echo (voice-to-text anywhere on your OS), Magic Beacon (highlight text in any app, trigger AI actions), Magic Tab (AI autocomplete in any app)

Why this matters for non-technical teams

The specific thing AnythingLLM solves that most AI tools don't: a non-coder can do the entire setup, and a non-coder can do the day-to-day administration.

Uploading documents is drag-and-drop. Creating a new workspace — a separate environment with its own documents, its own AI model, its own user access list — takes about 90 seconds in the UI. The Agent Flows builder is genuinely block-based; the only thing it resembles conceptually is a flowchart, not a programming language. Scheduled jobs are configured through a form, not a technical scheduling syntax.

For a legal team: a workspace per matter, documents dropped in as they arrive, an AI that answers questions with citations pointing to the exact clause in the exact filing. For a marketing team: brand guidelines, product docs, and competitive research all in one workspace, with an agent flow that can scrape a competitor's blog and summarize it on a schedule. For an ops team: internal runbooks indexed and queryable, with MCP connectors reaching out to Notion, Slack, or your ticketing system.

The privacy argument is real and specific here. Hosted AI products — ChatGPT, Claude.ai, Gemini — send your documents to their servers to process. For anything under NDA, attorney-client privilege, or basic competitive sensitivity, that's a policy question your legal team will have opinions about. AnythingLLM on a server your IT department controls means the documents stay inside your perimeter. The AI model calls go out to your configured provider (or don't, if you're running Ollama locally), but the document store never leaves.

Where it fits

AnythingLLM is not the only self-hosted AI workspace. Named alternatives worth comparing honestly:

• Open WebUI (91k stars) — excellent Ollama front-end, multi-user, actively maintained. Better for teams that just want a clean chat interface over local AI models; thinner on document search and has no visual agent flow builder. If your use case is "give the team a nice chat UI for Ollama," Open WebUI is cleaner. If you need workspaces, document intelligence, and no-code agents, AnythingLLM covers more ground.

• Flowise (35k stars) — a no-code AI app builder built on an open-source AI framework called LangChain. Deeper customization for building bespoke AI applications; steeper learning curve. AnythingLLM ships a product; Flowise gives you the tools to build one. They serve different personas.

• Khoj (28k stars) — personal AI assistant with strong document search and a well-regarded web interface. Closer in spirit to AnythingLLM; thinner on the agent builder and multi-user administration side. Good for solo use; AnythingLLM scales to teams more readily.

• PrivateGPT (55k stars) — the original "100% local, no internet" document-search project. No agent builder, no multi-user, fewer integrations. If your requirement is zero outbound calls to any outside service ever, PrivateGPT is the right answer. If you're comfortable with calls going to your chosen AI provider, AnythingLLM does significantly more.

Verdict

Install. For non-technical teams that need a private, document-aware AI workspace with an agent layer, there is no serious competitor at this price point (free).

The caveats are real but manageable. Multi-user and role-based access require the Docker deployment — the desktop app is single-user, which limits it to personal use. The Agent Flows builder is solid for linear workflows but not yet as expressive as something like Flowise for complex branching logic. And "bring your own API key" means AnythingLLM's monthly cost is zero, but your AI provider bill is not — factor that in before rolling it out to a large team on a cloud model.

None of those caveats disqualify it. The desktop app is a legitimate product for individuals and the server deployment is production-grade. The current version is stable, the project is well-funded (Mintplex Labs runs a cloud version that pays for development), and the MIT license means no licensing surprises if you embed this in a product or deploy it for a client.

If your team has been using a hosted AI product and someone has started asking the privacy question — AnythingLLM is where that conversation ends.

For your agent to install this

You can skip this section — it's the install details your AI agent needs. If you're using Cowork, Claude Code, ChatGPT Work, Codex, or Hermes, say "install AnythingLLM for me" and it will read the tables below.

System / OS requirements

API keys

Install commands

Desktop app (macOS / Windows / Linux) — recommended for individuals:

Navigate to anythingllm.com and download the installer for your OS. Run the installer. Open the app and follow the setup wizard to set your AI provider and first workspace. Setup takes roughly 10 minutes.

Docker (recommended for teams):

export STORAGE_LOCATION=$HOME/anythingllm && \
mkdir -p $STORAGE_LOCATION && \
touch "$STORAGE_LOCATION/.env" && \
docker run -d --rm -p 3001:3001 \
  --cap-add SYS_ADMIN \
  -v ${STORAGE_LOCATION}:/app/server/storage \
  -v ${STORAGE_LOCATION}/.env:/app/server/.env \
  -e STORAGE_DIR="/app/server/storage" \
  mintplexlabs/anythingllm

Then open http://localhost:3001 in a browser. The setup wizard walks through AI provider configuration, embedding model, and first admin account.

One-click cloud deploys:

AnythingLLM has official deploy buttons for AWS CloudFormation, Google Cloud Platform, DigitalOcean, Railway, Render, and Elestio — all linked from the README.

First-run checklist

1. Set AI provider + API key (or point to local Ollama instance)
2. Create first workspace
3. Drop in documents (PDF, DOCX, TXT, or paste a URL)
4. Start chatting — type @agent at the start of a message to invoke agent mode
5. To build an Agent Flow: Settings > Agent Flows > New Flow