What it is
AnythingLLM, from Mintplex Labs, bundles a chat interface, document ingestion and retrieval (RAG), a vector database, agents and MCP tools into one app. It runs as a single-user desktop app on Windows, macOS and Linux, or as a multi-user server in Docker with permissions, white-labelling and an embeddable website chat widget.
It works without extra services: a native embedder and the LanceDB vector store are built in, and a local engine based on Lemonade and llama.cpp runs models on CPU, Nvidia (CUDA) or AMD and Intel (Vulkan) GPUs. It also connects to Ollama, LM Studio and most cloud providers, and adds no-code agent flows, scheduled tasks, model routing and memories. Version 1.18.0 came out on 13 October 2026 under the MIT licence.
Our review
AnythingLLM removes nearly every step between a pile of documents and a chatbot that answers from them. The desktop app ships its own embedder, vector store and local model engine, so someone who has never heard of Ollama can install it, pick a model and drag in files. It has about 67,500 GitHub stars and released 1.18.0 on 13 October 2026.
The Docker edition is where it becomes a team product, with multiple users and permissions, white-labelling, a developer API and an embeddable chat widget. It is also where the risk sits. The repo lists 25 published security advisories, among them a high-severity remote code execution through argument injection in the filesystem-search agent skill (May 2026) and, in September 2026, a flaw that put the instance owner's private memories into the prompt seen by anonymous embed-widget visitors. Patch promptly and keep the server off the open internet.
Telemetry is on by default. The README lists what goes to PostHog: install type, document added or removed events, vector database type, LLM provider and model tag, and chat-sent events, with no content or IP address. DISABLE_TELEMETRY=true or the Privacy page turns it off, though the app still reaches cdn.anythingllm.com for model downloads and GitHub for some cached files.
Retrieval quality is the ceiling. Reviewers note that answers depend on chunking settings beginners rarely change, and the desktop edition stays single-user. Open WebUI gives finer admin control for large internal deployments; AnythingLLM's case is that a 10-person office can run it without an engineer, and hosted Cloud at $50 or $99 a month per instance is priced for that office rather than for one person.
Where it earns its place
An office manager at a 12-person accountancy
Let staff question engagement letters and internal procedures without uploading them to a public chatbot
Run the Docker edition on an office server with a local model through the built-in engine or Ollama, one workspace per team and per-user permissions. The software costs nothing under MIT; the real costs are a machine with enough memory for the model and someone to apply security updates.
A support lead at a SaaS company with a public docs site
Put a docs-trained chat widget on the help centre
Ingest the docs into a workspace and publish the embeddable widget from the Docker edition, backed by a cloud model. Hosted Cloud Basic at $50 a month avoids running the server; confirm the September 2026 embed-widget memory advisory is patched before going live.
A researcher with a laptop and a sensitive interview archive
Search and summarise hundreds of transcripts offline
Install the desktop app, keep the default embedder and LanceDB with a local model, and switch telemetry off under Privacy. Apart from model downloads from the AnythingLLM CDN, nothing leaves the laptop.
Before you commit
- Setup
- Desktop installers for Windows, macOS and Linux get to a first chat in minutes; the server is one Docker image, with deploy templates for AWS, GCP, DigitalOcean and Render and a bare-metal guide.
- Integrations
- Built-in engine, Ollama, LM Studio and most cloud LLM providers; LanceDB by default, PGVector and other vector stores; MCP tools, agent flows and scheduled tasks; a developer API and an embeddable widget.
- Data and privacy
- Local by default. Anonymous PostHog telemetry is on until DISABLE_TELEMETRY=true or the Privacy toggle; the README says chat and document content are never sent. 25 security advisories have been published, the most severe a May 2026 RCE in an agent skill.
- Cost at scale
- Self-hosting is free plus hardware and model costs. Hosted Cloud is $50 a month (Basic) or $99 a month (Pro, with a 72-hour support SLA) per private instance, with no per-seat price listed; Enterprise with SSO, RBAC and on-premise support is quoted.
- Limits
- Multi-user, white-labelling and the widget are Docker-only; the desktop app is single-user. Retrieval depends on chunking settings. About 330 open issues and a steady stream of security fixes.
What's good
- Built-in embedder, vector store and local model engine
- Desktop install with no other services to run
- Multi-user, white-labelling and embed widget in Docker
- MIT licence; telemetry documented and switchable
What's not
- Desktop edition is single-user
- Telemetry on by default
- 25 security advisories, one a high-severity RCE
- Retrieval quality needs chunking tuning
Best for
Individuals and small teams who want private chat over their own documents.
Alternatives
AnythingLLM: common questions
- Is AnythingLLM worth it?
- AnythingLLM is the shortest path from a folder of PDFs to a private chatbot, but 25 security advisories argue for keeping its server edition behind a login and a VPN. We rate it 4.3 out of 5.
- How much does AnythingLLM cost?
- Desktop and Docker free (MIT); an optional Desktop Pro subscription lifts daily limits on its Magic features (price not published in the docs). Hosted Cloud Basic $50/month, Pro $99/month; Enterprise custom.
- What are the downsides of AnythingLLM?
- Desktop edition is single-user; telemetry on by default; 25 security advisories, one a high-severity RCE; retrieval quality needs chunking tuning.
- Who is AnythingLLM best for?
- Individuals and small teams who want private chat over their own documents.
- How much does AnythingLLM cost for a team or in production?
- Self-hosting is free plus hardware and model costs. Hosted Cloud is $50 a month (Basic) or $99 a month (Pro, with a 72-hour support SLA) per private instance, with no per-seat price listed; Enterprise with SSO, RBAC and on-premise support is quoted.
- Is AnythingLLM safe for company data?
- Local by default. Anonymous PostHog telemetry is on until DISABLE_TELEMETRY=true or the Privacy toggle; the README says chat and document content are never sent. 25 security advisories have been published, the most severe a May 2026 RCE in an agent skill.
- What are the best AnythingLLM alternatives?
- The closest alternatives we have reviewed are Open WebUI, Jan and LibreChat.
Answers checked Oct 23, 2026.
Make AnythingLLM? Show this rating on your site
