AnythingLLM, built by YCombinator-backed Mintplex Labs and first released in 2023, is an open-source (MIT) "all-in-one" AI app centered on document RAG and agents. It runs as a native desktop app or a self-hosted Docker container, requires no code, and connects to more than 30 LLM providers — local options like Ollama, LM Studio and LocalAI, plus the major clouds — alongside a wide range of vector databases. Its workspace model keeps separate chat contexts, each with its own model, embeddings and prompt settings, and it supports the Model Context Protocol for connecting external tools.
Beyond retrieval, AnythingLLM includes a no-code agent builder with built-in skills (web browsing, scraping, SQL, file operations), visual Agent Flows, and scheduled agent tasks. The project is active — roughly 62.7k GitHub stars and frequent releases (v1.15.0, the latest, shipped in mid-2026) — and recent additions include dynamic model routing, Intelligent Tool Selection (only relevant tools/MCPs are injected into the prompt), a beta Community Hub for sharing system prompts, skills and flows, and the v1.15.0 "Magic Features" that bring on-device dictation, text actions and autocomplete to any app. The managed cloud now spans Basic ($50/mo), Pro ($99/mo) and Enterprise. The desktop app is a fairly heavy Electron build, hosted cloud tiers are comparatively expensive, and as always the quality of RAG answers depends on the model you point it at. For a local-first knowledge assistant, though, it's one of the most complete options.
Key Benefits
- Local-first knowledge base: Turn your documents into a private, queryable assistant with no cloud requirement.
- Bring any model: Broad provider support means you're never locked into one vendor.
- Agents without code: Build tool-using agents and schedule recurring tasks from a GUI.
- Deploy your way: Desktop, Docker or managed cloud, all from the same project.
Use Cases
- Chat with your documents — Index company or personal files for grounded, cited answers.
- Private research assistant — Combine local models with RAG for sensitive material.
- No-code agents — Automate web research, scraping or database queries with built-in skills.
- Team knowledge hub — Self-host a multi-user instance with per-workspace settings.