temuxllm

Local LLM on Android, ollama-style. LiteRT-LM 0.11.0-rc1 + Gemma 4 + Termux client. 127.0.0.1 only.

A self-contained Android app + Termux client that runs Google's Gemma 4 (or Qwen3) on your phone's GPU and exposes it as http://127.0.0.1:11434/api/generate. No network. No cloud. No data leaves the device.

Use cases / 用途 No network or cloud: Run LLM prompts on airplane mode, in subways, in sensitive contexts where queries cannot leave the device. Ollama-compatible endpoint: Apps and scripts expecting http://127.0.0.1:11434 (Ollama's standard port) work without modification. Termux scripting: Pipe LLM output into shell pipelines: curl -s ... | jq .response for structured queries on the phone. Offline assistant: Gemma-class model resident in memory, sub-second replies for short queries. Interchangeable models: Load any .litertlm model (Gemma, Qwen, future releases) and swap freely. Private prototyping: Build against a local endpoint during early dev — no API keys, no token costs, no prompts sent to third parties.