dalao.ai

AnythingLLM

A local-first AI app that puts document chat, AI agents, scheduled jobs and a meeting assistant in one program; repository MIT-licensed, desktop installer under separate proprietary termsOpen source

About AnythingLLM

AnythingLLM, from Mintplex Labs, is a local-first AI app that puts document chat, AI agents, scheduled jobs and a meeting assistant in one program, with local or cloud models. It ships as a desktop app, a Docker image and an Android app, plus a hosted Cloud. As of 2026-10-26 the latest GitHub release is v1.18.0, published 2026-10-14 00:15 (UTC+8), which swaps the built-in engine from Ollama to AMD's Lemonade with llama.cpp. The repository is MIT-licensed, the desktop build carries separate proprietary terms, and Cloud starts at $50 a month.

EditorialUpdated 7 min read

What problem it solves

Hooking internal documents up to a large model usually means stitching together a model, a vector database, document parsing, a chat UI and permissions. AnythingLLM packs these into one app. Documents such as PDF, TXT and DOCX go into a workspace, and chats over them come with citations. The same interface adds AI agents, MCP support, scheduled jobs and custom skills. Models can run on the built-in local engine or come from cloud providers.

The site pitches it as a private AI assistant that runs entirely on the local computer, with no accounts, no API keys and no token limits.

Forms:

  • Desktop app: native builds for macOS on Apple Silicon and Intel, Windows on x64 and ARM, and Linux.
  • Docker self-hosting: multi-user support, permissions, white-labeling and an embeddable chat widget, which the README marks as Docker-only.
  • Android app: a separate anythingllm-mobile repository under GPL-3.0.
  • Cloud: hosted multi-user instances run by the vendor.

The repository is Mintplex-Labs/anything-llm, created 2023-06-04 10:29 (UTC+8), with master as its main branch. Recent releases: v1.16.1 on 2026-08-28 00:42 (UTC+8), v1.16.2 on 2026-09-23 03:06 (UTC+8), v1.17.0 on 2026-10-02 06:17 (UTC+8) and v1.18.0 on 2026-10-14 00:15 (UTC+8).

The publisher is Mintplex Labs, Inc., the copyright holder in LICENSE and the contracting party in the desktop terms of use. According to the YC company page, AnythingLLM is in the Summer 2022 batch, with founder Timothy Carambat, described as Founder of Mintplex Labs Inc, and a team size of 4.

AnythingLLM is designed as a local-first application.

AnythingLLM self-hosted terms

Core capabilities

A new built-in engine. v1.18.0 replaces the desktop app's bundled Ollama with Lemond, the embedded build of AMD's open-source Lemonade server, running llama.cpp underneath. On AMD Ryzen AI machines it can use the NPU through FastFlowLM, and on Snapdragon devices it uses GenieX. The vendor blog says new models and performance work land in llama.cpp first, and that upgrading the bundled Ollama kept breaking features. Existing users get an engine upgrade prompt; the upgrade removes the old engine and the models downloaded with it, leaving workspaces, chats and documents intact.

Reasoning controls. v1.18.0 adds reasoning effort or thinking toggles to the chat input for local and cloud models, shown only when the app can detect what a model supports.

Documents and retrieval. Answers within a workspace cite their sources. The document library is separate from workspaces: removing a document from a workspace deletes only that workspace's vectors, while deleting it from the library also clears the parsed file and cached embeddings.

Agents and automation. The README lists a no-code agent builder, MCP compatibility, custom agents, dynamic model routing, automatic and user-managed memories, and cron-scheduled jobs.

System-wide desktop features. The site lists a meeting assistant with no bots joining the meeting and no cloud processing; v1.17.0 rebuilt it on NVIDIA Nemotron 3 Diarization and Moondream Parakeet Redux. The Magic features cover dictation with Magic Echo, highlight-and-ask with Magic Beacon and autocomplete with Magic Tab, each with a daily free allowance.

Mobile. The Android app runs small models on the phone or connects over the local network to LM Studio, llama.cpp, Ollama or the desktop app.

Who uses it

The site cites 67k+ GitHub stars, 200+ contributors and 5M+ Docker pulls. The Y Combinator page says more than 2.6M people use AnythingLLM every month across Desktop and Docker, calling that a floor because most users turn telemetry off, and the desktop privacy policy cites more than 1M downloads (as reported by the vendor). As of 2026-10-26 the GitHub API shows 67,550 stars and 7,586 forks; contributors number 271 by GitHub account, or 277 including anonymous ones.

For organizations, the Cloud page offers multi-user instances with tenant isolation and white-labeling, and the Enterprise tier adds on-premise deployment, SSO and role-based access control.

How it compares

Open WebUI is also a self-hostable chat interface, but its license adds a branding clause to a BSD-3 base: except for deployments with 50 or fewer end users in a rolling 30-day period, with prior written permission, or under an enterprise license, altering or removing the Open WebUI branding is a material breach. LM Studio is a closed-source desktop app under its own app terms, focused on downloading and running local models. Jan is an open-source desktop app under Apache-2.0.

The AnythingLLM repository is MIT-licensed and can be modified and white-labeled freely. It bundles document chat, agents, scheduled jobs and multi-user permissions in one app. Its desktop build carries separate proprietary terms.

Who it's for / Not for

A good fit for:

  • Individuals and teams who want document chat and AI agents on their own devices or servers. The self-hosted terms state it can run fully air-gapped with local models and a local vector database.
  • Teams that need multiple users, permissions and white-labeling, which the Docker build provides; Cloud also offers multi-user instances with tenant isolation.
  • Desktop users who want local models to work out of the box. The built-in engine needs no separate inference server, and the system requirements page lists 16GB of RAM and an 8-core CPU for a basic setup.

A poor fit for:

  • Organizations that need fine-grained permissions. Issue 1787, requesting fine-grained access controls, opened 2024-06-29 04:03 (UTC+8) and remains open as of 2026-10-26.
  • Self-hosted deployments that need a formal SLA. The self-hosted terms state the standard version comes without one unless a separate enterprise agreement is signed.
  • Linux desktop users who want every Pro feature: the Pro docs say all features are available on Windows and macOS and do not mention Linux.

Getting started

Desktop. The download page offers an installer per system that needs no account. New installs run the Lemonade engine, and a small model matched to the machine can be downloaded on first launch; cloud provider API keys can be added in settings instead.

Docker. The official quickstart gives a Docker command, after which the app is at http://localhost:3001. Adding -e ENABLE_BUILT_IN_LLM=true turns on the built-in engine inside the container, with NVIDIA, AMD and Intel GPU support. Data persists across container rebuilds and new image pulls. A pg image connects to PostgreSQL with PGVector.

Telemetry. Setting DISABLE_TELEMETRY to true in the server or Docker .env turns it off, as does the Privacy option in the app sidebar.

Developers running from source get a monorepo with six parts: frontend, server, collector, docker, embed and browser-extension. The yarn setup command creates the .env files for each part.

Pricing & license

Self-hosting and the desktop app are free; Cloud and Pro are paid:

  • Cloud Basic: $50 a month, with a customer-supplied LLM API key.
  • Cloud Pro: $99 a month, with a 72-hour support SLA.
  • Enterprise: by quote, with on-premise deployment, a custom SLA, SSO and role-based access control.
  • Desktop Pro: lifts the daily limits on Magic Echo, Magic Beacon and Magic Tab and removes AnythingLLM branding from generated documents. A license key goes into settings after purchase, the license is validated periodically, and a lapsed subscription falls back to the free allowances. The vendor states that whatever is free today stays free.

License: the repository LICENSE is The MIT License, copyright Mintplex Labs Inc., and the self-hosted terms say the AnythingLLM core is provided under MIT. The desktop terms of use, however, grant a non-exclusive license for personal or internal business purposes and bar modifying, adapting or creating derivative works. The Android app repository is GPL-3.0.

Risks & limits

On privacy, the official sources say the following: the download page says Nothing phones home; the README describes telemetry that collects anonymous usage data and is turned off through DISABLE_TELEMETRY or the in-app Privacy setting. The README also lists outbound connections that remain with telemetry off: configured external model, embedding or vector database providers, the cdn.anythingllm.com model mirror, and GitHub-hosted files used for context window caching. Air-gapped setups need the embedder and reranker models supplied by hand.

On licensing, the official sources say the following: the site and the self-hosted terms state MIT; the desktop terms of use call the software's structure and source code trade secrets, limit use to personal or internal business purposes and bar derivative works.

Cloud limits. Hosted instances have no GPU and no built-in LLM; the built-in embedder can crash an instance with a 502 on very large documents; custom agents and MCP are not supported for security reasons. The Cloud terms say user data stays owned by the account owner but grant usage rights for platform improvement, and the vendor may cancel a subscription for any reason at any time; data is not retained after cancellation. The limitations page names the tiers Starter and Professional, while the pricing page uses Basic and Pro.

Numbers. The Y Combinator page cites 56K+ stars, the site 67k+, and the GitHub API 67,550. The 2.6M monthly users and 1M+ downloads are as reported by the vendor.

Engine migration. The v1.18.0 upgrade deletes models downloaded with Ollama, which must be fetched again for the new engine, and the changelog says a future update will migrate users automatically.

Our verdict

What sets AnythingLLM apart: document chat, AI agents, scheduled jobs and a meeting assistant in one app; a built-in local engine in the desktop app that needs no separate inference server; an MIT-licensed repository, with multi-user support, permissions and white-labeling in the Docker build; and Cloud from $50 a month.

Its limits: the desktop installer comes under proprietary terms of use, unlike the MIT-licensed repository; outbound connections such as the model mirror remain with telemetry off; the request for fine-grained permissions (issue 1787) is still open; and the standard self-hosted version comes without an SLA unless a separate enterprise agreement is signed.

Sources

Official

  1. AnythingLLM home pageanythingllm.com · Accessed Oct 26, 2026
  2. Download AnythingLLManythingllm.com · Accessed Oct 26, 2026
  3. AnythingLLM Cloudanythingllm.com · Accessed Oct 26, 2026
  4. AnythingLLM Mobileanythingllm.com · Accessed Oct 26, 2026
  5. Why AnythingLLM is moving from Ollama to Lemonade and llama.cppanythingllm.com · Accessed Oct 26, 2026
  6. AnythingLLM Docs homedocs.anythingllm.com · Accessed Oct 26, 2026
  7. Why AnythingLLM Prodocs.anythingllm.com · Accessed Oct 26, 2026
  8. Getting Your Pro Keydocs.anythingllm.com · Accessed Oct 26, 2026

Updates & reading