Models on this phone (#63)
Settings → Agents and models → Models on this phone: download a model and Buddy runs it on the phone itself. Nothing leaves the phone, and it works in airplane mode.
- Gemma 4 E2B, Gemma 4 E4B, Qwen3.5 2B and Qwen3.5 4B (all Apache-2.0), each checked against its checksum
- The engine is llama.cpp's official Android build, fetched with the first model (about 33 MB once installed)
- Runs on the NPU (Snapdragon 8 Gen 2 and newer), the GPU or the CPU; Auto takes the fastest that starts
- Pick the model in an OpenCode chat, under On this phone. Buddy starts it when the chat needs it, stops it after 10 minutes without use, and stops it when the phone gets too hot
- On a OnePlus 12, Gemma 4 E2B on the NPU reads about 2,000 tokens/s and writes about 24/s. Small models handle simple requests (the battery, a timer, a quick look at an app) and make mistakes on longer tasks
Lite mode
For small models: the agent starts with a short prompt and a dozen phone tools, and loads more with more_tools when the task needs them. The first request goes from about 24k tokens to about 4k. On by itself for models on this phone, and a switch in Chat options for any OpenCode chat.
Fixes
- OpenCode 2.3 and later got Buddy's instructions twice in every chat
- With background mode, Android's app installer could open on the agent's hidden screen, where you couldn't tap it
- Background chats paused when you touched your own screen
Full Changelog: v2.3.13...v2.3.14
Which file? Download buddy.apk: it updates itself. Buddy is also on F-Droid, where each version arrives a few days after it is released here. fdroid-check-only-v2.3.14.apk is only there for F-Droid's reproducible-build check, and doesn't update itself.