AI assistant

GrowRig includes an optional AI assistant — a chat that answers questions with the context of a specific grow. It is built on the same integrations system as everything else external, so the model provider is pluggable rather than baked in.

How it’s powered

The assistant consumes the ai.chat capability (and ai.vision for images). Any integration that provides those capabilities can back it — the seed bundle is Ollama, which runs against a local or hosted Ollama-compatible API, so conversations can stay entirely on your own hardware.

Because the provider is chosen by a binding, you can point the assistant at a local model globally and override it per grow, without the feature knowing or caring which model answers.

Grow-scoped conversations

A conversation is owned by a user and scoped to one grow. Each chat records the grow and environment it belongs to, so the assistant can answer with the relevant cultivation and climate context rather than in a vacuum.

Conversations are persisted: they have titles, a running message count, a preview, and can be archived. Every message is an immutable user or assistant turn, kept in order — so a chat is a durable record you can return to, not a throwaway session.

What it’s for

The assistant is a helper layered on top of the grow-domain model — for interpreting readings, sanity-checking a plan, or answering “what should I look for at this stage?” It does not command equipment; automation stays with the reconciliation engine, whose decisions are always explainable on their own. Keeping the assistant advisory means an AI provider being slow, offline, or absent never affects climate control.

Enabling it

Configure an AI integration instance (for example, Ollama) under Control panel → Integrations and confirm its connection test passes.
Bind the ai.chat capability to that instance — globally, or for a specific grow.
Open a grow and start a conversation.

If no ai.chat binding resolves, the assistant is simply unavailable; the rest of GrowRig is unaffected.