AI Data Visualization Starter

An AI data analyst powered by Runtype, deployed to InsForge,
with generative charts using Flint.

You bring:

• A Postgres database with business data (sample data included)
• A question about the data

You get:

• A relevant dashboard, table and SQL (generates charts using Flint, Microsoft's AI charting library)
• Scheduled monitoring that re-checks the same data over time and flags what slipped

It works across:

• Web (using persona.js, Runtype's agent UI library)
• Slack
• API

Access to the web is gated behind auth, so only users you want have access to query data.

Start here: Runtype configuration

The entire product is composed in one file: runtype.config.json.

It defines:

• Two agents — the conversational Business Analyst and the Daily Business Monitor
• Product surfaces — the chat surface (with WebMCP interactive charts), a Slack surface (same agent, chart images inline), and an API surface
• Scheduling — the monitor's cron schedule and timezone
• Integrations and wiring — the InsForge Postgres tools, chart rendering, per-user records, and an analyst-grounding eval suite

Inspect and validate it locally with the CLI, then restore it into your own Runtype account:

# validate the config locally (no account needed)
npx @runtypelabs/cli validate-product runtype.config.json

# restore it: open the dashboard template importer pointed at the config
open "https://use.runtype.com/now?templateUrl=https://raw.githubusercontent.com/runtypelabs/ai-data-visualization/main/runtype.config.json"

The importer prompts for the template variables (product name, your InsForge base URL, business context, monitor schedule) and the one pending secret, INSFORGE_API_KEY. The same URL works with Runtype's MCP tooling (create_product_from_example with a url).

The runnable app implementation lives in ./insforge — it's the supporting half that serves the web app and hosts the data.

Architecture

flowchart LR
  Config[runtype.config.json] --> Analyst[Business Analyst agent]
  Config --> Monitor[Scheduled daily monitor]
  Config --> Chat[Chat surface + WebMCP charts]
  Config --> Slack[Slack surface + chart images]
  Config --> API[API surface]
  App[insforge/ app] --> Chat
  Analyst --> DB[(InsForge Postgres)]
  Analyst --> Chat
  Analyst --> Slack
  Monitor --> DB
  Monitor --> Records[(Snapshot + alert records)]

Loading

Runtype supplies the intelligence: schema discovery, inspectable text-to-SQL, generated dashboards, actions, monitoring. InsForge supplies the backend — Postgres, auth, edge functions serving the web app. Point the template at a different database and the same config becomes an analyst for whatever your data tracks: the agents discover your schema live, nothing is hard-coded.

Working on browser-tool latency? See the Runtype local-tool resume analysis for the reproduced loop, measurements, app mitigation, and recommended runtime regression test.

Deploy the full demo

1. The Runtype half — restore runtype.config.json as above.

2. The InsForge half — seed the sample dataset and deploy the web app:

export INSFORGE_BASE_URL=https://<your-app>.insforge.app INSFORGE_ADMIN_KEY=<key>
node scripts/seed-sample-data.mjs        # optional sample dataset (skip for your own data)
node scripts/deploy-flint-render.mjs     # chart-image renderer for Slack/scheduled surfaces
RUNTYPE_CLIENT_TOKEN=ct_live_... INSFORGE_ANON_KEY=anon_... node scripts/deploy-shell.mjs

Your analyst is live at https://<your-app>.insforge.app/functions/ai-data-visualization. Full setup — including end-user auth with per-user row-level security — is documented in insforge/README.md. Working with a coding agent? SKILL.md is an agent-consumable install guide for the whole thing.

License

Apache 2.0 — see LICENSE.