Self-hosted & local AI
Run real AI on hardware you already own.
LowSpec AI tracks which AI tools actually run on a $200 mini PC, an old laptop or a 4 GB NAS — with the licence, RAM floor and power draw checked by hand, not copied from a vendor blog.
Most worth your watts
Ranked by how well they hold up on low-power hardware — not by who paid us.
Immich
Google Photos replacement with on-device AI search. Face recognition and CLIP-powered search run on your own CPU — no cloud, no API key.
llama.cpp
The C++ engine underneath most local inference. Ugly flags, best-in-class CPU performance, runs a 1B model in 2 GB of RAM.
Ollama
The default way to run a local LLM. One binary, a model library, and an OpenAI-compatible endpoint on port 11434.
Open WebUI
A self-hosted ChatGPT-style interface that sits on top of Ollama or any OpenAI-compatible API. Accounts, history, RAG and document upload included.
Paperless-ngx
Scan-and-forget document archive with local OCR. Drops a paperless household on 2 GB of RAM and gives you full-text search over every invoice you own.
whisper.cpp
OpenAI's Whisper speech-to-text reimplemented in C++. Transcribes a podcast on a CPU with roughly 1 GB of RAM and no Python.
Buying guides
How these numbers are collected
Every tool page carries a verified line with a date and a source:
| Source | What it means |
|---|---|
| hands-on | We installed it and measured RAM, throughput or power ourselves. Date = the day we ran it. |
| docs | Taken from official documentation. No measurement, so treat it as a floor, not a promise. |
| vendor-claim | The vendor's own marketing number. Unverified — we say so on the page. |