The Keep loop
Report, Script, Watch, Dashboard, App — the five shapes an answer can take, and when to reach for each.
Most AI data tools give you an answer and forget it. Tablize turns good answers into assets that keep working while you’re not looking.
There are five shapes an answer can take. Every kept asset is one of these. You pick the shape from the Keep bar at the bottom of the answer — no form, no category picker, no “save as” modal.
The rest of this page walks through each shape in a little more depth — what it does, when to reach for it, what it costs you.
Report — a written answer
A Report is the most common way to keep an answer. It’s a Markdown document with charts baked in. The Agent fills it out from the answer you just got, but you can edit anything.
When to use it: the answer is something you want to read later — a weekly digest, a client-facing recap, a post-mortem. If somebody will scan it on a phone, it’s a Report.
What you get:
- A Markdown page with charts, tables, and KPI cards.
- A shareable link that works for anyone you send it to (no Tablize account needed).
- A schedule — e.g. “re-run this every Monday at 9:00”. The new run gets its own dated version.
- A comparison view — this week vs last week, diffed automatically.
Cost: negligible. Reports are cached Markdown + charts; rerunning one uses the same tokens as asking the question fresh.
See: Reports.
Script — rerunnable logic
A Script is the code behind an answer, saved for reuse. It’s the side you don’t see in the chat: the SQL, the Python, the data-joining, the munging.
When to use it: you want to run the same analysis on different data. A consultant doing cohort retention for twelve clients a month. A dev comparing this sprint’s error rate to last sprint’s. Anything where the question is constant and the data isn’t.
What you get:
- A named, parameterized piece of code.
- A one-click “run on new data” — the Agent auto-maps the new schema onto the script’s expected columns.
- Version history. Forking. Scheduled runs.
Cost: small. You pay for the run (the tokens + compute), not for keeping the Script.
See: Scripts.
Watch — a recurring check
A Watch is a Script that runs on a schedule and only speaks up when something is off. It is silent by default. You don’t look at a Watch; it looks at your data on your behalf.
When to use it: you care about a threshold — refund rate above 3%, freezer above −15°C, new-user signups more than two standard deviations below trend. Anything where “no news is good news.”
What you get:
- A schedule and a condition.
- A quiet hours window.
- Notification routing — Slack, email, webhook.
- An acknowledge + snooze flow when it does fire.
Cost: small, and only when the check runs. Watches on large datasets can batch — ask the Agent if you’re not sure.
See: Watches.
Dashboard — a live view
A Dashboard is what you build when multiple answers belong on one page, updated live, shown to multiple people.
When to use it: your team asks the same three or four questions every day and you’re tired of re-running them. Or you want one page to drop into a Notion doc for the leadership meeting.
What you get:
- Pinned live panels (each one is a mini-Report that auto-refreshes).
- Multiple themes (monochrome, arctic, synthwave, 11 in total).
- A public link you can embed anywhere.
- The same CRUD guarantees as Apps — dashboard components can read from tables live.
Cost: depends on refresh frequency. A live sensor dashboard costs more than a daily finance one.
See: Dashboards.
App — a mini-tool for teammates
An App is the heaviest Keep shape. It’s a small web application the Agent generates — HTML, CSS, JS, and CRUD endpoints wired directly to your tables.
When to use it: the answer isn’t “here’s what I found” but “here’s a tool the team needs.” A return-request form. An inventory editor. A customer-lookup page for support.
What you get:
- Generated UI with full CRUD against one or more tables.
- A Data Contract — the App declares which queries it needs; the backend serves them directly from Postgres, milliseconds-fast, no Agent in the loop.
- Public or team-only sharing.
- Version rollback to any previous generation.
- Theme switching — same 11 themes as Dashboards.
Cost: generation uses tokens. After that, Apps are free to serve (CRUD goes direct to Postgres).
See: Apps.
Which shape when?
When you’re not sure, use this:
| If the answer… | Reach for… |
|---|---|
| is something you want to read later | Report |
| is the same question on different data each time | Script |
| is a threshold you want to be alerted about | Watch |
| is multiple answers shown to multiple people live | Dashboard |
| is a tool a teammate needs to use without asking you | App |
You can also keep the same answer in multiple shapes. A churn analysis can become a Report (the executive recap), a Script (rerun on next month), and a Watch (tell me if churn spikes) — all three from one conversation.
Next steps
Read the dedicated docs for each shape when you need depth:
Or, if you haven’t actually tried the loop yet, run through the Quickstart first.