DD-008
Compo
Parents tutoring grade-school writing
Drop a model essay + topic; print an A4 writing guide your kid can actually follow.
Reference repoSoyaOS gets out of your way. Bring your own LLM. Deploy any Agent as an OpenAI-compatible model โ locally, on your cluster, or hosted.
MIT-licensed ยท no telemetry ยท runs on macOS / Linux / Windows
$curl -L https://soyaos.ai/install | sh
$soyaos start
# paste 127.0.0.1:7474/v1 into Cherry Studio.
# done.
One process. Speaks OpenAI on
/v1/chat/completions. Loads a SoyaPack, schedules a Comet, talks to your model of choice.
Four flagship stories
The alpha ships with four reference Agents โ each a complete SoyaPack you can run, fork, or read for inspiration.
DD-008
Parents tutoring grade-school writing
Drop a model essay + topic; print an A4 writing guide your kid can actually follow.
Reference repoDD-009
Knowledge workers tracking a beat
Type "watch USB-C policy in EU" in Chinese โ get a long-image digest in DingTalk at 9am daily.
Reference repoDD-010
Real-estate self-media operators
One sentence in โ 500-row Excel out; each row has a "make post / make reel" button that just works.
Reference repoDD-011
Solo short-video creators
Natural-language brief โ Remotion timeline โ Comet renders MP4 โ lands on your NAS overnight.
Reference repoWhy SoyaOS
Three properties we refuse to compromise on, even as the alpha ships weekly.
Solo on a laptop to Enterprise on-prem. Same kernel, same SoyaPack manifest format, different deployment shapes. You graduate without rewriting.
Talk to SoyaOS through any OpenAI SDK. SoyaOS talks to any OpenAI-compatible upstream. Bring your own LLM โ Claude, GPT, Qwen, a local Ollama โ and swap freely.
Capability tokens, signed SoyaPacks, an end-to-end audit trail in SoyaScope. No telemetry, no phone-home โ your agents stay on your hardware unless you say otherwise.
Ready when you are
Planet (control) ยท Moon (UI / edge) ยท Comet (worker). Start with all three on your laptop, peel them apart as you grow.