Cactus Chat

Cactus Compute

About this app

Offline, private AI.

Cactus Chat lets you talk to AI, directly on your phone. This means it's free, runs offline, and your data stays on your device.

If you're a developer, use Cactus to benchmark the latency and throughput of various LLMs.

Ratings and reviews

It's a decent enough demo but isn't nearly as fast as the GitHub description says it should be, I'm only getting about 9 tokens a second on a Pixel 7 with the default model. If this wasn't open source it would be rated much lower, there are applications such as PocketPal using a 1b gemma3 model which is as fast or faster in a feature rich application with a smarter LLM model.

Running gemma3:1b on my Oneplus 11, the latency with Cactus Chat was less than 1 second, while the latency with Ollama was more than 10 seconds.

it exactly do what it does thanks, i would like to see a benchmarking tool to test llms and a new feature to import from phone storage would be appreciated 👏