Flows and AI FAQ

Answers about automation and AI in Hark.

What is the difference between a chat flow and the AI bot?

A chat flow is a scripted tree you build node by node: it does exactly what you designed, every time. The AI bot answers freely using your connected model and knowledge. A chat widget runs in one mode or the other, and flows can call AI nodes for the parts where a model helps, so you can mix scripted certainty with AI flexibility.

What happens if my AI provider is down?

Hark degrades honestly. An AI flow node takes its failure branch and hands off to a human, the bot escalates rather than guessing, and semantic search falls back to keyword search. The AI never fabricates an answer to cover an outage.

Will publishing a flow change break conversations already in progress?

No. A run walks the exact tree it started on, snapshotted at start. Publishing a new version only affects conversations that begin after the publish.

Can I require review before a flow goes live?

Yes. Turn on approval for the flow, and publishing creates a pending approval holding the proposed tree; a reviewer approves it to publish or rejects it. You can also schedule a publish for a future time.

Do I need to bring my own AI provider?

Yes. You connect a provider (OpenAI, Anthropic, an OpenAI-compatible endpoint, LM Studio, or Ollama) and choose your chat and embedding models, so you control the model and the cost. See AI configuration and providers.

What can the copilot do without my confirmation?

It can read your workspace within your permissions and answer questions. For any change (closing a conversation, assigning it, publishing an article, and so on) it proposes the action and waits for you to confirm; it never performs a write on its own.