Mintplex-Labs/anything-llm

What it adds beyond the summary

The summary lists features but omits the feature/deployment split and the operational details an engineer needs before adopting. Three things matter here.

Feature parity depends on deployment mode. Several headline capabilities are Docker-only: multi-user instance support and permissioning, and the custom embeddable chat widget for websites. The desktop app (Mac/Windows/Linux) does not carry these. If you need multi-user or embedding, the desktop build is not a path — you must self-host via Docker.

Token-cost claim. "Intelligent Skill Selection" is stated to reduce token usage by up to 80% per query while keeping unlimited tools enabled for a model. This is the page's only quantified performance claim and it's a vendor figure with no baseline or setting given — treat it as marketing until verified.

How to use it

For a from-source dev setup (not Docker), the page gives the exact sequence, run from repo root:

yarn setup

Then fill in the .env files in each application section — the page explicitly warns that server/.env.development must be filled "or else things won't work right." Then:

yarn dev:server
yarn dev:frontend
yarn dev:collector

The monorepo has six sections: frontend (Vite + React), server (NodeJS express — handles vectorDB and LLM interactions), collector (NodeJS express — parses documents), docker, embed (submodule), and browser-extension (submodule).

Limits

• Telemetry is on by default. It collects anonymous events (install type, document add/remove, vector DB type, LLM provider/model tag, chat-sent events — no content) via PostHog. Opt out with DISABLE_TELEMETRY="true" in the server/docker .env, or in-app under Privacy.
• Outbound connections persist even with telemetry disabled: cdn.anythingllm.com (model mirror CDN), github/githubusercontent.com (flat files for context-window caching), plus any external LLM/embedder/vectorDB provider you configure. If you need a fully air-gapped deployment, audit these.
• STT support is listed as Native Browser Built-in only — no external speech-to-text providers, unlike the broad TTS provider list (OpenAI, ElevenLabs, PiperTTS, etc.). Verify this if voice input is a requirement.