Agent modes overview
Every AI agent in Userorbit operates in a mode that determines how much initiative it takes. Choosing the right mode lets you balance speed against control — start hands-on, then loosen the reins as you gain confidence in the agent's output.
Assistant mode
In assistant mode the agent is entirely reactive. It waits for you to open the chat interface and ask a question, request a draft, or trigger a quick action. No background processing occurs, and nothing is published or sent without your direct instruction.
Use assistant mode when you want full control over every interaction — for example, during the first week of using agents, or when working on sensitive communications like incident updates.
Supervised mode
In supervised mode the agent actively monitors events in your workspace — new feedback submissions, scheduled intervals, or other triggers you configure. When it detects a relevant event, it generates a proposed action (a draft response, a categorization suggestion, a content update) and places it in the agent inbox.
A team member must review each proposal and explicitly approve or reject it before anything is executed. This mode is ideal for teams that want agents to do the heavy lifting of drafting and triaging while keeping a human in the loop for quality control.
Autonomous mode
Autonomous mode gives the agent permission to act on its own — but only when its confidence score meets or exceeds the threshold you set. Proposals that fall below the threshold are still routed to the inbox for human review.
This mode works best for high-volume, well-understood tasks like categorizing feedback into existing buckets or sending templated acknowledgment responses where the risk of an incorrect action is low.
Confidence thresholds
Every action an agent considers is assigned a confidence score between 0 and 100. The score reflects how certain the agent is that the proposed action is correct given the available context.
- High threshold (80–100) — Only the most clear-cut actions execute automatically. Most proposals go to the inbox. Good for getting started.
- Medium threshold (50–79) — A balanced setting. Routine actions execute automatically while ambiguous cases get human review.
- Low threshold (20–49) — The agent handles most tasks on its own. Use this only after you have reviewed enough proposals to trust the agent's judgment for this particular workflow.
You can adjust the threshold per agent and per workflow, so a feedback-categorization agent might run at a lower threshold than a public-announcement agent.
Capabilities by module
Agent capabilities vary depending on which Userorbit module they interact with:
- Help center — Draft new articles, suggest edits to existing ones, flag stale content.
- Feedback — Categorize submissions, detect duplicates, draft responses, summarize trends.
- Announcements — Generate drafts from release notes or changelogs, suggest audience segments.
- Changelog — Create entries from commit messages, pull requests, or manual prompts.
Choosing the right mode
A practical approach is to start every new agent in assistant mode, move to supervised mode once you are comfortable with the quality of its suggestions, and finally enable autonomous mode for the tasks where accuracy is consistently high. You can change the mode at any time from the agent settings page without losing any existing configuration.