TO DO Smartplant

✅ Done in v3.0

• Kernel with a real plugin API (use, registerPlugin, getSensor, analyze)
• Sensor abstraction layer + 7 drivers (mock, manual, serial, mqtt, http, homeassistant, electrode)
• Persistent plant memory with care log, species profile and 1.x migration
• Multi-provider AI with retries, caching, structured output and offline fallback
• First-person voice with 5 personas across 10 languages
• Semantic event bus (plant:thirsty, plant:stressed, …)
• All 10 plugins rewritten on the kernel, each with a passing test
• CLI split from the library, with one-shot commands for scripts and cron
• Electrophysiology, vision, symbolic knowledge and semantic memory
• Symbiosis layers: fusion, control, safety, evidence, personalization
• Hardware autodetection, firmware generation, OpenClaw adapter

☀️ Core

• TypeScript declaration files (.d.ts) generated from the JSDoc
• Reading downsampling — keep minute resolution for a day, hourly beyond that, so long-lived plants don't grow an unbounded memory file
• Actuator drivers: plant.body.control decides what to do and safety approves it, but executing it still means writing your own execute(). A driver contract for pumps, valves, smart plugs and motors is the missing half.
• Multi-plant manager: one process, one memory root, many plants

🌡️ Sensors

• Bluetooth driver for Xiaomi Mi Flora / HHCC (very common, very cheap)
• Camera input — leaf colour and wilting as additional signals via a vision model
• Sensor calibration offsets per driver

🧠 AI

• Embed the species profile once and reuse it, instead of resending it each call
• Token accounting exposed on plant.ai.stats
• Streaming replies for speak()

💡 Ideas

• Web dashboard reading the memory JSON directly
• Export to Home Assistant as a set of entities, closing the loop both ways
• Community species-profile registry so the first AI call is unnecessary