Start here
What AgenticObjects is, how App and Central Management Console fit together, and the fastest path from your first brief to a number you can stand behind
AgenticObjects gives your company analytics agents that answer business questions with numbers you can stand behind. Every figure arrives with where it came from, how far it can be trusted and the query that produced it — so you can act on an answer, not just read it.
What AgenticObjects does
You ask an agent a question in plain language. The agent maps your words onto a governed data model your company has built in Model Studio, runs queries against your own data source, and writes back an answer with its evidence attached.
Agents also work while you are away. On a schedule, an agent studies your data and publishes a brief: a short written report of what it found and what it recommends, delivered to the people it is meant for.
Three things make those answers dependable:
- Every number carries a trust tier. You can see at a glance whether a figure is a governed metric, built from governed metrics, or exploratory. See Trust tiers.
- Every agent works inside declared boundaries. It reaches only the data model, metrics and rows it has been given. See What an agent can see.
- Every run keeps its own record. Steps, duration, cost and the query behind each number are recorded as the run happens.
App and Central Management Console
AgenticObjects is one web application with two shells. The one you see depends on your role, and both read the same records.
App — ask and read
The App is where business users work every day.
- Dashboard — what needs your attention today: the latest brief, open recommendations and findings. Home
- Chats — ask a question and get an answer you can check. Chat
- Agents — the agents available to you; start a conversation from any card. Agents
- Briefs — everything delivered to you, with the runs that produced it. Briefs & runs
- Timeline — which run produced what, in order. Timeline
Central Management Console — set up and govern
The Central Management Console is where administrators shape everything the App shows: agents and their instructions, connections and the live data models, users, roles and row-level security, schedules, budgets and spend, and the audit record. Good places to begin are Agent Studio, Data models and Users & Roles.
Your first hour
Take these steps in order — each one builds on the last.
- Read a brief. It is the quickest way to see what an agent does for you. Reading a brief
- Open a finding. Follow a number back to its evidence. Finding detail
- Ask a follow-up in chat. Ask about what the brief covered, then about what it did not. Chat
- Learn the trust tiers. Once you can read the badge beside a number, you know how much weight it can carry. Trust tiers
- Know an agent's boundaries. They explain why two people can see different totals for the same question. What an agent can see
Setting AgenticObjects up for your company? Start with Connections and Data models, then create an agent in Agent Studio and keep an eye on Usage and spend.
Tip: When a number surprises you, read the assumptions printed above it before anything else: the period, the scope, and the metric the agent matched your words to.
Principles you can rely on
Data models are made in Model Studio. Your governed vocabulary — metrics, dimensions and synonyms — is built in Model Studio, or imported there from a model file, and goes live from it. In the Central Management Console you bind a model to a connection and to agents, choose which metrics each agent may use, and set row-level security.
Nothing is ever deleted. Agents, models, rules, users, projects and workspaces are archived. An archived item leaves your working lists, keeps its full history, and can be brought back with its contents intact.
The record cannot be rewritten. The audit record shows who changed what, and when. It is read-only for everyone, administrators included. See Audit.
People choose the model. The language model behind an agent changes only when an administrator changes it, and that change is recorded.
How this guide is organized
Every article says at the top which shell it belongs to. There are two kinds:
| Kind | What it gives you |
|---|---|
| Concept | How something works and how to read it — this article, Trust tiers, What an agent can see |
| Screen guide | A walk through one screen: what is on it, what each part is for, and what to do next |
Read the concepts first. The screen guides build on them, which keeps each one short and to the point.