Data overview and Agent Copilot
An agent's Data overview explains available material and possible directions, not verified business findings. It only shows sources currently authorized for that agent. Opening it does not grant access to unselected workspace data or sibling datasets on a connection.
Overview versus shared profile
Switch sources inside the agent to inspect meanings, possible uses and limitations. This changes the displayed overview, not agent permissions or the recommendation scope. Authorized managers maintain shared field meanings in the Data Mart profile. Members and shared recipients do not gain management rights by viewing an overview.
A shared profile is jointly maintained source metadata. An agent overview may also use your background and your own query experience to form a private interpretation. Neither is a shared fact or evidence for quoting numbers or source text. Actual tasks must still read authorized sources.
Preparation, reuse and retry
The system checks current access and matching cached understanding first. An ordinary overview request can automatically schedule missing understanding for the selected source. Switching sources may require waiting. Completed matching understanding can be reused; opening the panel does not always recompute it. Source definitions, permissions, language or personal background can require a different version.
Preparation shows its status. Use explicit generate/retry controls after a generation failure. For a network read failure, retry loading first; a successful reload is not proof of successful re-analysis. Overview preparation, chat analysis and new recommendation generation are distinct operations, not universally free or manual. See account usage.
Choose the action you mean
- Add background: describe the subject, purpose and limits. This is caller-private chat background, not automatic workspace consensus or a completed data query.
- Analyze directly: send a concrete task. It reads evidence within current agent authority and follows normal metering.
- Click a recommended or preset question: run the full question directly, without appending an old draft. Simple/deeper groups do not change your selected analysis depth.
- Discuss a suggestion: clarify a direction or proposed improvement with Copilot. Discussing, viewing or ignoring is not execution or approval of shared changes.
- Save personal knowledge: use an explicit save action. Mentioning context in chat does not automatically create a pinned personal fact.
- Confirm a shared change: review the proposed wording and scope and have an authorized person explicitly confirm it. Private exploration and chat are not automatically published as consensus.
Recommendation lifecycle
The current version checks authorization before restoring a usable batch for the current user, agent, session, language and chat surface. Prepared batches are reused; failed reads cannot fall back to unverified old cards. Only an explicit New questions action requests a new metered batch. Opening guidance, discussing Copilot advice or ordinary data refreshes does not authorize replacement.
Initial scope preparation can take time. When an old batch cannot be safely restored or no cards remain, follow the interface rather than expect automatic refill. Recommendation scope focuses questions without expanding chat authority; the overview's selected source does not change it. This is not a promise that all initial preparation is free or that every restart environment was tested.
Two different Copilots
Data onboarding Copilot helps managers explore, validate and save external sources. Agent Copilot suggests work using this agent's authorized material and your context. It does not automatically open connections, change shared definitions or authorize background automation.
Next: MCP onboarding, Knowledge, Verification.