← Back to all projects

AnythingLLM

A self-hosted ChatGPT over your own documents, with no-code agents — an app you run, not a library you build on.

OfficialMIT
Stars
67.8k
Forks
7.6k
Open issues
348
Last commit
30 Oct 2026

What is AnythingLLM?

AnythingLLM ships the whole retrieval stack as one installable application: a document collector, an embedding model, a vector database and the chat interface arrive together, so you drop files into a workspace and start asking questions without wiring services to each other. Documents attached to a chat go into the model's context in full by default, and only when they overflow the context window does the app offer to chunk and embed them, after which answers are assembled from a few retrieved snippets. New workspaces run in agent mode, so the same chat window can browse the web, query a SQL database, call MCP servers or run a flow you drew in the visual editor. The trade-off is that this is an application rather than a library — you extend it through its settings, its custom-skill plugin format and its REST API, never by importing it into a program of your own — and the three ways to run it are not the same product: the desktop build has no accounts, the Docker build has no bundled model, and the hosted tier has neither MCP nor custom skills.

What can you do with AnythingLLM?

  • Attached documents go in whole; embedding is the fallback — Drop a file into a chat and its full text is placed in the context window. Only when that overflows does AnythingLLM offer to chunk and embed it, and it warns that continuing anyway lets it prune the context to fit, losing data.
  • Where a document lives decides who can read it — A file attached inside one thread is unavailable in the next one. Embedding it instead publishes it to every thread in that workspace — and, on a multi-user instance, to every account with access to the workspace.
  • Retrieval is tuned per workspace, not once for the instance — Each workspace carries its own system prompt, model and vector settings: how many snippets to send, which the documentation suggests keeping at four to six for most models, and a similarity threshold that discards matches under 20% by default. An accuracy-optimised mode reranks results for roughly 100-500 ms more, but only on the bundled LanceDB.
  • New workspaces are agentic before you ask — Agent mode has been the default for new workspaces since v1.11.1, so a chat can already browse and scrape the web, query a SQL database, draw charts, write files and reach Gmail, Outlook or Google Calendar. Where the mode is off, typing @agent opens an agentic session for that conversation.
  • Extend the agent with drawn flows or MCP servers — Agent Flows chain five block types — web scraper, API call, LLM instruction, read file, write file — into a skill the model picks like any built-in one. MCP servers declared in anythingllm_mcp_servers.json boot on demand and land in the same tool list.
  • Put an agent on a cron schedule — Scheduled Jobs, added in v1.13.0, run an agent unattended with the skills, flows and MCP servers you pick for it, and show the fifty most recent runs of each job for review. One run executes at a time by default, and anything still going after five minutes is stopped.

Before you choose AnythingLLM

  • Switching a self-hosted instance to multi-user mode cannot be undone, and as of v1.16.0 it also removes Scheduled Jobs, which the documentation states is available in single-user mode only.
  • Mintplex Labs states that its hosted cloud disables MCP servers and custom-coded agent skills for security reasons, and it ships no built-in model, leaving that tier the thinnest of the three.

Star history

17 Aug to 30 Oct · +2.9k

64.8k67.8k

Frequently asked questions

Is AnythingLLM free for commercial use?

AnythingLLM is released under the MIT licence — OSI-approved open source, which permits commercial use.

How can AnythingLLM be deployed?

AnythingLLM is available as Self-hosted / Runs locally / Managed cloud.