SHSelfHostIndex

Self-Hosted AI

Run large language models, chat UIs, and machine-learning tools locally with self-hosted AI software.

137 self-hosted apps · 328 comparisons

All self-hosted ai apps

Last reviewed Aug 26, 2026 · 454 words

The 138 tools filed under "self-hosted AI" are really four shelves in one aisle: engines that run model weights, chat frontends for talking to them, visual builders for assembling LLM apps, and agents that go off and do things on your behalf. Most bad installs in this category are shelf errors — someone wanted a private ChatGPT and ended up deploying a workflow orchestrator with a Postgres dependency. Figure out which shelf you are on and the shortlist mostly picks itself.

How to choose across the four shelves

Three axes do the splitting. First, the layer: if you want to chat, you need a runner from Local LLM Runners plus a frontend from AI Chat UIs — nothing in this parent category is that. Second, where the intelligence actually lives: a surprising share of "self-hosted AI", including the biggest agents here, is orchestration around a paid API key, so your prompts still leave the building unless you deliberately point the tool at local models. Third, hardware: agent shells idle at 512 MB, app platforms like Dify want 4 GB, and anything that loads model weights itself starts at 8 GB and climbs from there.

Where to start

For teams building LLM features, Dify (Apache-2.0, 153,524 stars) bundles a visual workflow builder, RAG pipelines, and observability in one multi-container stack — heavier than a toy, but it replaces three or four separate tools. For a coding agent, OpenHands (MIT, 4 GB) writes code, runs commands, and browses the web inside a sandboxed Docker runtime, and works against local or remote models; budget for token costs if you use a hosted provider.

Big star counts, small amounts of local AI

OpenClaw tops this category at 387,634 stars, and it is a genuinely useful assistant — inbox, calendar, flight check-ins, driven from WhatsApp or Telegram. It is also the clearest example of the category's asterisk: it requires a model API key, ships a headless browser with real reach into your accounts, and its own docs stress that security needs careful setup. The hottest projects here are agents orchestrating cloud models, not local inference. That is a fine trade, but make it on purpose, and give any agent the least access you can get away with.

Where I'd start

Run a small model locally with a runner and a chat UI first — it is a weekend project and teaches you the RAM and latency realities everything else in this category inherits. Once that works, my default second install is Dify: broad enough to cover RAG, agents, and workflows, and boring enough (Apache-2.0, Docker Compose) to still be running next year.

Curated picks

Easiest Self-Hosted AI to self-hostBest API Self-Hosted AIBest Beginner-Friendly Self-Hosted AIBest Cross-Platform Self-Hosted AIBest Docker Self-Hosted AIBest Free Self-Hosted AIBest Kubernetes Self-Hosted AIBest Lightweight Self-Hosted AIBest Multi-User Self-Hosted AIBest No-Database Self-Hosted AIBest Offline-First Self-Hosted AIBest Plugins Self-Hosted AIBest PostgreSQL Self-Hosted AIBest Real-Time Self-Hosted AIBest Single-User Self-Hosted AI

Self-Hosted AI comparisons

328 head-to-head comparisons in this category.

See all comparisons →