About AnythingLLM Desktop 1.16.1 — Private AI Chat With Your Documents
AnythingLLM is an all-in-one AI application from Mintplex Labs. You connect a model, local or hosted, add your documents, and chat with them, with agents, memory and scheduled tasks built in. It comes as a desktop app for Windows, macOS and Linux, and as a Docker server for teams. It is MIT licensed.
This page links to the official 1.16.1 desktop builds on GitHub. Bineret did not build AnythingLLM and is not affiliated with Mintplex Labs.
What it does
- Chat with your documents. PDF, TXT, DOCX and more, with drag-and-drop uploads and citations pointing back to the source.
- Local by default. The desktop app runs models locally through its own engine, with LanceDB as the built-in vector database, so documents and chats can stay on your machine.
- Or any provider. OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock, Mistral, Groq, DeepSeek, OpenRouter, Ollama, LM Studio and many more.
- Agents. A no-code agent builder, MCP compatibility, custom agents, and agents that can browse the web.
- Memory and schedules. Automatic and user-managed memories, and scheduled tasks that run prompts on a cron schedule with full agent abilities.
- Model routing. Rules decide which provider and model handle each chat.
- A developer API for building on top of it.
Desktop or Docker
The download here is the desktop app, for one person on one computer. Two features are Docker only: multiple users with permissions, and the chat widget you can embed on a website. For a team, run the Docker image or one of the cloud templates instead.
Telemetry is on by default
AnythingLLM sends anonymous usage events to PostHog unless you turn it off. The project lists exactly what: the install type, that a document was added or removed, the vector database and model provider in use, and that a chat was sent, with no content and, it says, no IP address. Switch it off under Privacy in the sidebar, or set DISABLE_TELEMETRY=true on a server. Even then the app contacts the model and database providers you configure, its own model CDN, and GitHub for some cached files.
Free, with an optional Pro tier
Since 1.15.0 there is AnythingLLM Pro. Mintplex Labs describes it as purely additive: nothing existing is locked away, and each Pro feature keeps a free daily allowance with no sign-up. The first Pro features, “Magic Features” such as on-device dictation that works in any app, are free up to a daily limit.
What 1.16.1 changed
Released on 27 August 2026. Agent reasoning and tool calls now fold into one collapsible block. Leaving a chat mid-answer shows a warning, since leaving cancels the response. On Windows it embeds Microsoft’s Foundry Local engine, for Windows 11 24H2 and later, and on Snapdragon X devices a new GenieX engine replaces the old NPU engine. Models from the old Snapdragon engine are deleted on update and must be downloaded again. The Windows installer also gains silent install and uninstall flags.
Which file to download
The builds are large because they bundle a local inference engine:
- Windows x64:
AnythingLLMDesktop.exe, 391.5 MB. Windows on ARM: 497.2 MB. - macOS Apple Silicon: 522.2 MB. macOS Intel: 489.9 MB.
- Linux x64 AppImage: 860.9 MB. Linux ARM64 AppImage: 601.8 MB.
When to use something else
If a team will share it, you want the Docker version, not this download. If you only chat with hosted models and never with documents, a lighter chat client will do. If you are building workflows rather than chatting, a visual builder such as Langflow is the better fit.
Key capabilities
Your documents, your machine
Drop in PDFs, text and Word files and ask questions with citations back to the source. With the built-in local model and the default LanceDB store, neither the documents nor the chat have to leave your computer.
Any model you like
The local engine, or OpenAI, Anthropic, Gemini, Azure, Bedrock, Mistral, Groq, DeepSeek, OpenRouter, Ollama, LM Studio and dozens more, with rules that route each chat to a chosen provider and model.
Agents that keep working
Build agents without code, connect MCP tools, give them memories, and schedule prompts to run on a cron schedule with full agent abilities. In 1.16.1 their reasoning and tool calls fold into one collapsible block.
Desktop is for one person
Multi-user accounts with permissions and the embeddable website chat widget exist only in the Docker version. For a team, run Docker rather than installing this desktop build on a shared machine.
Telemetry, stated plainly
Anonymous usage events go to PostHog by default: install type, document added or removed, providers in use, and that a chat was sent, never its content. Turn it off under Privacy in the sidebar.
Snapdragon users, note this
1.16.1 replaces the Snapdragon NPU engine with GenieX and deletes models from the old engine on update, so plan to download them again. Windows 11 24H2 and later also get Microsoft Foundry Local built in.
Questions & Answers
Is AnythingLLM free?
Yes, MIT licensed. An optional Pro tier adds features on top; Mintplex Labs says nothing existing is locked behind it and every Pro feature keeps a free daily allowance without sign-up.
Do my documents leave my computer?
Not if you use the built-in local model and the default LanceDB vector store. If you choose a hosted model or database, the text you send goes to that provider.
Does it collect data about me?
It sends anonymous usage events by default: install type, document added or removed, vector database and model provider in use, and that a chat was sent, without content. Turn it off under Privacy in the sidebar.
Can several people use it?
Not the desktop app. Multi-user support with permissions is only in the Docker version.
Why is the download so big?
The desktop builds bundle a local inference engine so you can run models without installing anything else. The Linux x64 AppImage is the largest at 860.9 MB.
I use a Snapdragon laptop. Anything to know about 1.16.1?
Yes. A new GenieX engine replaces the old Snapdragon NPU engine, and models downloaded for the old engine are deleted on update. You will need to download them again. The project recommends GGUF models on this engine, since NPU-only models have 4K context windows.
Is Bineret connected to AnythingLLM?
No. This page describes AnythingLLM and links to the builds Mintplex Labs publishes on GitHub.
Tutorials
Pick the build for your computer
Windows on a normal Intel or AMD PC: AnythingLLMDesktop.exe. Windows on ARM, such as Snapdragon X laptops: AnythingLLMDesktop-Arm64.exe. A Mac with an M-series chip: AnythingLLMDesktop-Silicon.dmg; an Intel Mac: AnythingLLMDesktop.dmg. Linux: the AppImage for your architecture.
Install
On Windows run the installer; on macOS open the disk image and drag the app into Applications. On Linux:
chmod +x AnythingLLMDesktop.AppImage
./AnythingLLMDesktop.AppImageThe release also carries installer.sh, a script that installs the Linux build for you.
Silent install on Windows
1.16.1 adds silent and headless install and uninstall flags for managed deployments. The exact flags are in the Windows installation docs.
First run
Choose a model: the built-in local engine, or a hosted provider with its API key. Create a workspace, drop documents into it, and ask questions; answers cite the documents they came from.
Turn telemetry off if you want
Anonymous usage telemetry is on by default. Open Privacy in the sidebar and disable it.
For a team: Docker instead
Multi-user accounts and the embeddable website chat widget are only in the Docker version. Follow HOW_TO_USE_DOCKER.md in the repository, and set DISABLE_TELEMETRY=true in its .env to opt out there.
Support
Where support comes from
Mintplex Labs maintains AnythingLLM. Start with the documentation, report bugs as GitHub issues, and use the project's Discord community for questions.
What Bineret covers
This page. If a download link is broken or something written here is wrong, tell us and we will fix it. We do not provide models, API keys, hosting or set-up help for AnythingLLM.
Paid features
AnythingLLM Pro and the hosted instance are sold by Mintplex Labs, not by Bineret. Questions about them go to Mintplex Labs.
No warranty
The MIT licence provides the software as-is. If your workspaces matter, back up the app's storage folder before updating.
