Skip to main content
Research and ops teams spend hours pulling the same threads: what changed this week, what themes repeat in customer feedback, and what leadership should see before a meeting. Manual copy-paste from docs, tickets, and spreadsheets does not scale. UtopikAI lets you connect those sources once, then run an agent on a schedule to produce a structured summary — without someone sitting in chat each time. Use an assistant when a person needs to explore and follow up interactively. Use an agent for recurring digests, trend reports, or workflow-driven research. They are two different entities; each has its own models, knowledge, skills, tools, and prompts.

​
A research agent

Create a custom agent for research or reporting. Write instructions that define:
  • What sources to prioritize
  • How to structure the output (sections, bullets, citations)
  • What to do when data is missing or conflicting
Attach the document sets and tools the agent should use. Assign skills for repeatable analysis steps, for example a competitive scan or a weekly metrics template.

​
Knowledge worth connecting

Typical sources, via connectors:
  • Product docs and release notes
  • Customer feedback and support history
  • Internal wikis and strategy decks
  • Spreadsheets or exports your team already maintains
Connectors are optional. Index them when you want answers grounded in those sources. UtopikAI keeps connected knowledge in sync on each connector’s refresh schedule.

​
Scheduled tasks

Attach tasks to the agent so runs happen without a live chat session. From the agent setup wizard or Task Management, create a task with:
  • Instructions — What to research and how to format the result
  • Schedule type — Single run or recurring
  • Input Files — Optional files the agent should read each run
See Tasks Overview and Creating Agents for the full form. Example jobs:
  • Weekly summary of new product questions from support
  • Monthly competitive feature comparison from connected docs
  • Daily digest of status-page incidents and runbook updates

​
Workflows for review before delivery

When a person must approve the report before it is shared, run the agent inside a workflow:
  1. Agent gathers and drafts the summary
  2. A user approval step sends the draft to a reviewer
  3. A follow-up step posts or saves the approved version
Workflows are automations on a canvas. They are not the same as Studio, which is for generating images, video, audio, and 3D.

​
Assistants for ad hoc follow-up

Keep a separate assistant for interactive drill-down. After a scheduled report lands, Workspace members can ask in chat:
  • Why did this metric spike?
  • Show the source doc for that claim.
  • Compare this month to last quarter.
Do not use a chat assistant as a substitute for unattended runs. Create an agent for that.