Run AI on your own machine. Pick your hardware, install one tool, chat with a model that fits — this page gets a beginner there in about 15 minutes.
Backstory · Follow updates on X
Localist is a beginner-first guide, not another giant tool list. You start from the machine you already own, follow one guide, and skip the other 50 options. And it doesn't rot: a pipeline — not vibes — refreshes the Fresh updates section every morning, and we review and prune stale recommendations instead of hoarding them.
Jump to: Start here · The essentials · Fresh updates · Glossary · Contributing
Don't read everything. Find your row, follow one guide, and you'll be chatting with a local model in under 15 minutes.
| I have… | Your guide |
|---|---|
| An NVIDIA GPU (any GeForce RTX) | NVIDIA GPU path |
| A Mac with Apple Silicon (M-series) | Mac path |
| An AMD GPU (Radeon RX) | AMD GPU path |
| Just a laptop / no GPU | CPU-only path |
| A mini PC or home server | CPU-only path — or the NVIDIA path if it has a GPU |
| No idea what I have | Start here |
Everything here can run locally; many paths work offline after setup, and your prompts stay on your machine unless you connect a cloud service.
One opinionated pick per category. Alternatives are inside each entry — but if you're new, just take the pick and move on.
Which tool first? Want clicks? Start with LM Studio. Comfortable with one command? Start with Ollama. You do not need both.
| Category | Our pick | Why this one |
|---|---|---|
| Model runner | Ollama | One command to install, one to run a model. The de-facto beginner standard. |
| Desktop app | LM Studio | Point-and-click everything: browse, download, and chat with models. No terminal needed. |
| Web UI | Open WebUI | ChatGPT-style interface on top of Ollama. Multi-user, chat with your documents, voice. |
| Chat models | Qwen3 family | Strong at every size, tiny to huge. Which size fits you → choosing models. |
| Coding models | Qwen Coder | Same idea, meaningfully better at code. Pairs with the coding agent below. |
| Coding agent | OpenCode | Most popular open-source coding agent; points at your local models. |
| Image generation | ComfyUI | Node-based, runs every major open image model. Steeper curve, unmatched power. |
| Chat with your docs | AnythingLLM | Point it at a folder and a local model — done. |
| Speech-to-text | whisper.cpp | Fast local transcription on any hardware. |
| Text-to-speech | Piper | Fast, natural offline voices — runs even on a Raspberry Pi. |
When you outgrow beginner tools, llama.cpp is the engine worth learning — it's what most of the tools above are built on.
The full curated set (with licenses, hardware fit, and honest caveats) lives in
data/curated.yml — it's the single source of truth these
picks come from.
New projects and tool releases from the past week, refreshed daily by the pipeline. Pipeline-surfaced projects are not curated recommendations. Start with the Essentials table above if you want the trusted beginner picks.
Updated 2026-07-26
🆕 New & active projects
- lidge-jun/opencodex — Universal provider proxy for OpenAI Codex & Claude Code — use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama…) with Codex CLI, App, SDK, and Claude Code · ⭐ 709
- drumih/turbo-fieldfare — Gemma 4 26B-A4B inference in ~2 GB of RAM on any M-series MacBook · ⭐ 67
📦 Tool releases
- LM Studio 0.4.20
- Jan v0.8.4 — Migration Settings and credentials now live in a backend-managed store.
- Ollama v0.32.3 — Fixed model downloads that stall before sending data. Improved integrations: restored Claude Code Channels, fixed Anthropic thinking streams, and made Hermes Desktop respect
--force-build. - Ollama v0.32.4 — Support Laguna on Apple GPUs via the MLX engine Quantize draft-model output heads at the requested type when creating speculative-decoding drafts.
- vLLM v0.26.0 — vLLM v0.26.0 Release Notes This release features 411 commits from 212 contributors (61 new)!
The glossary explains every term you'll bump into — GGUF, quantization, context window, VRAM, tokens/sec — in plain words. And choosing models answers the #1 question: which model size actually fits my machine?
Found a great tool? Spotted a dead project? Open an issue — suggest a tool · report a stale entry. See CONTRIBUTING.md for the ground rules.