Device & Cloud Collaboration
An agentic TuyaOpen device splits the work in two: the device does what must be instant, the cloud does what must be smart. Pick a recipe below to watch them collaborate.
See it work: a smart oven
Choose a recipe, then press Run to step through the collaboration.
Press Run to start the recipe.
Cloud agent ยท reasoning
โ
Plan: 180ยฐC, ~12 min, both elements
โ
Decide to check near the end
โ
Assess doneness from the image
Device ยท tool calls
โ
set_temperature(180)Set the oven to 180ยฐCโ
set_heat(both, on)Turn on top and bottom heatโ
set_time(12)Set a 12-minute timerโ
start()Start bakingโ
take_photo()Photograph the cookiesโ
stop()Golden โ stop the ovenWho does what
The same split applies to every agentic device, not just the oven. Real-time perception and actuation live on the device; knowledge, reasoning, and language live in the cloud.
โ
On the device
Real-time ยท local ยท low-latency
- Capture microphone audio and detect speech (VAD)
- Wake word and chat modes โ decide when to listen
- Play the spoken reply, music, and prompt tones
- Capture camera frames and show a live preview
- Render the on-screen chat UI
- Run MCP tool calls on real sensors and actuators
- Stay responsive with low latency, even on a flaky link
MCPโ
โ
In the cloud
Knowledge ยท reasoning ยท always improving
- Speech recognition (ASR) โ speech to text
- Language understanding and generation (NLG)
- World knowledge and step-by-step reasoning
- Agent roles, prompts, and long-term memory
- Skill orchestration and task planning
- Cloud MCP to external services and APIs
- Multilingual voices โ improves without re-flashing
Device handles what must be instant. Cloud handles what must be smart.