NeuroFunky

For self-hosted vs hosted, I tried to make it easy to set up, but I'm wondering about teams that don’t want to deal with infrastructure setup and maintenance. By the way, you're totally right about privacy with the hosted version.

About the model not being confident about a recommendation: for now, there is no real mechanism to handle this, but automations can be set up with human-in-the-loop approval. That’s also something that can be addressed with prompting, but yes, I probably need to think about a smarter way to handle it

thanks! right now each automation execution is stateless from the AI perspective so each run is independant. btw there is prompt variable injection from event payloads to provide state and context.

this is an interesting topic, I have a few ideas on how to do this, maybe via a summary of recent executions injected into the prompt context