AnythingLLM

Yes. The AnythingLLM desktop app is completely free and open source for macOS, Windows, and Linux — no account, no subscription, no feature gating. Self-hosting via Docker is also free. The only paid tier is AnythingLLM Cloud (managed hosting) which starts around $50/month for teams that don't want to run their own infrastructure. For individuals and small teams willing to use their own machine or server, AnythingLLM is one of the best zero-cost AI workspaces available.

Yes, if you pair it with a local LLM. AnythingLLM ships with built-in support for Ollama, LM Studio, and LocalAI, which let you run models like Llama 3, Qwen, Mistral, and Gemma entirely on your own hardware. Once your model is downloaded and your documents are ingested, you can chat with your files with zero network access — ideal for confidential data and air-gapped environments.

AnythingLLM ingests PDF, DOCX, TXT, CSV, Markdown, HTML, EPUB, images (with OCR), and audio files (with transcription). It also supports scraping websites directly into a workspace, importing Confluence spaces, GitHub repos, and YouTube transcripts. Each workspace maintains its own vector database so retrieval stays scoped — a common problem with simpler RAG tools that dump everything into one global index.

Ollama and LM Studio are LLM runners — they load and serve models but don't handle documents, users, or RAG out of the box. AnythingLLM sits on top and adds workspaces, vector storage, document ingestion, multi-user accounts, and agent tooling. Most local-AI users actually run both: Ollama as the model engine, AnythingLLM as the chat and document interface.

The app itself is lightweight. The hardware requirements come from whichever LLM you pair it with. For cloud models (OpenAI, Claude) any laptop works. For small local models (7B Llama 3) you want 16GB RAM minimum. For 13B-70B models 32GB-64GB RAM is recommended, and a modern GPU dramatically improves response time. Apple Silicon Macs are particularly well-suited thanks to unified memory.

Yes. AnythingLLM supports multi-user workspaces, role-based permissions, shared vector databases, and embeddable chat widgets. Teams commonly deploy it on a shared server or via AnythingLLM Cloud ($50+/month) and give each employee accounts with access to specific document workspaces. It's a popular choice for legal, finance, and healthcare teams that need private RAG without sending data to OpenAI or Anthropic.

Yes. The platform includes built-in agent tooling with web browsing, code execution, file system access, SQL querying, and custom skills. You can chain tools together and let the agent work through multi-step tasks across your documents and the web. While not as advanced as dedicated agent frameworks, it's enough for most knowledge-worker automations — research, report generation, data lookup, and document summarization.

AnythingLLM is built by Mintplex Labs, a US-based open-source AI startup. The project is hosted on GitHub under a permissive license, with thousands of stars and an active contributor community. The company's business model is the managed AnythingLLM Cloud tier plus enterprise support contracts — the desktop and self-hosted versions remain free forever, which is a deliberate positioning choice against closed-source competitors.