Core concepts

If you read one page before the rest of the docs, read this one. Nothing here is technical, but every other page assumes you know it.

The Agent is a control plane

In most AI products, the LLM is a helper — a smart autocomplete glued to a button. In Tablize, the Agent is the entire control surface. It sees your question, picks which tools to call, runs them in sequence, reads the results, and narrates the answer back to you.

That means there’s no “SQL mode” vs “chart mode” vs “automation mode”. You ask one question and the Agent decides which of its ~100 tools apply. If the question is a lookup, it writes SQL. If it’s a modeling question, it reaches for Python. If you ask it to monitor something, it creates a Watch.

You can always see what it did — tool calls are rows in the chat — and you can always edit them. But you don’t have to pre-plan the plumbing.

Workspaces and sessions

A workspace is your Tablize account. It contains:

• Your uploaded tables, connected databases, and integrations.
• Every session you’ve ever opened (a session = a chat conversation).
• Every asset you kept — Reports, Scripts, Watches, Dashboards, Apps.
• Your team, if you have one.

Sessions are cheap. Open a new one whenever you shift topics. The Agent starts each session with a clean memory but full access to the workspace, so any table you uploaded last week is still there.

Kept assets — Reports, Scripts, etc. — live at the workspace level, not the session. Once you save a Report, it doesn’t disappear when you close the session.

Four states of the layout

Tablize has one window, but it slides between four states depending on what the Agent is doing.

State	When you're in it	What it looks like
Chat	Default — you just asked a question that doesn't produce a big artifact.	Chat centered, 720px wide. Like ChatGPT.
Split	The Agent produced something worth looking at — a report, a generated app, a map.	Chat on the left, panel on the right. The panel slides in on its own.
Artifact	You want to look at the result full-screen — reviewing a long report, testing a generated app.	Panel takes the whole window. A floating 💬 button returns you to chat.
Flow	You want to see everything this session produced at a glance.	An Artifact Navigator — a time-ordered DAG of every tool call and output.

You don’t choose states from a menu. The Agent picks them based on what it just produced, and you can override any choice with a single click.

Creators and Viewers

Workspaces have two kinds of seats:

• Creators can chat with the Agent, connect data, build assets. They consume tokens from the plan’s rolling allowance.
• Viewers cannot chat. They can open any Report or Dashboard that’s been shared with them, click around, read, export. They don’t consume tokens.

This split matters for billing. A team of three analysts and eight readers costs you three Creator seats plus eight Viewer seats at $5/month each — not eleven Creator seats.

Viewers are also how you share an answer externally without giving anyone an agent. Drop a link, let them read.

The Keep language

Once an answer is worth keeping, Tablize gives you five shapes to keep it in:

• Report — a written answer. Markdown with charts. Shareable link. Re-runs on a schedule.
• Script — the code behind an answer, rerunnable on new data.
• Watch — a silent check that pings you only when something moves.
• Dashboard — a live view, sharable with teammates or embeddable.
• App — a small CRUD tool for non-technical teammates.

The full tour is on The Keep loop. For now, the important bit is that every good answer has a “Keep bar” under it offering the shapes that make sense. You don’t open a form; you pick a shape.

Five domains under the hood

You don’t need to know this to use the product, but it helps when reading later docs.

The Agent’s ~100 tools are organized into five domains:

• Data — tables, SQL, Python, ETL, analysis.
• IoT — MQTT devices, sensors, cameras, space-asset-point modeling.
• App — generating mini-apps and dashboards.
• Media — file storage for things that aren’t tables (images, PDFs).
• Platform — auth, billing, scheduling, governance.

Plus:

• Integrations — 39 connectors like Stripe, Shopify, GA4.

These are invisible in normal use — the Agent just calls the right tool. They become visible when you’re debugging (every tool call names its domain) or when you’re choosing a plan, since plans gate access by domain.

Next steps

Now that the mental model is in place, the rest of the docs will make sense in any order.

• Quickstart — sign up and try the loop.
• The Keep loop — all five ways to keep an answer.
• Asking questions — how to phrase things for the Agent.