Q: Do I need to tell my agent exactly where to find information, or can it figure that out on its own?

A: It depends on the proactive workflow you've set up. For any integration you've connected (for example Gmail), you define the instructions, and the agent then works autonomously from that data based on what you've told it to watch for.

Q: What actually makes ChatChat's proactive agents different from a typical AI copilot?

A: A few things together: a Living Brain that gives agents real long-term memory and context, so they surface connections you might otherwise miss; a wider integration set; smart model routing for better efficiency; and proactive workflows that are event-driven rather than only scheduled, meaning an agent can message you first when something happens, like a new email landing, not just at a set time each day.

You can also chat with real friends and family alongside your agents in the same group chats, which most copilots don't support.

Q: What kinds of work should I automate first?

A: Start with repeated information gathering, summarisation, preparation or coordination tasks—for example morning briefings, meeting preparation, follow-up drafts, weekly reports, monitoring approved sources or recurring checklists. Run the workflow manually first. Convert it into a scheduled task only after the manual result is accurate and consistently useful.

Q: What makes a scheduled workflow reliable?

A: Use a narrow and measurable objective, approved and current sources, clear date ranges, a defined agent role, only the necessary connectors, a structured output format and explicit escalation rules. Test the workflow manually before scheduling it, then review early results for errors, outdated sources or unnecessary tasks.

Q: How should I configure a deep-research task and reduce fabricated sources?

A: Define the research question, date range and acceptable source types. Ask the agent to break the work into sub-questions, prioritise primary sources, provide exact titles, publishers, dates and URLs, and distinguish facts from estimates or interpretation. Require important claims to be cross-checked and every URL to be opened or verified. Treat inaccessible or missing sources as unverified rather than allowing the agent to invent a citation.

Q: How can I stop an email-monitoring agent from reporting promotions instead of bills?

A: Use Gmail filters or labels together with a narrow scheduled-task prompt. Create a label for bills using known senders, domains or subjects, then tell the agent to search only that label and a defined time period. Put global behaviour in the agent instructions, the recurring report criteria in the scheduled task, and deterministic inclusion or exclusion rules in Gmail itself. Test the search manually before scheduling it.