Clara Insights

Insights is an analytics studio inside Kibu Documentation. You ask Clara, Kibu’s AI assistant, a question about your data in plain English, and you get back a table you can keep refining in conversation.

What it does

Ask a question in plain English and get a table back.
Refine the table conversationally, then view it as a chart or a summary.
Export to CSV, save it, or put it on a schedule.
Hit Explain to get a plain-English summary of any Insight.
Start from a template gallery of ready-made reports — MAR, billing, documentation compliance, demographics and more.

Under the hood

Every Insight is one table backed by one query, written in KQL — Kibu Query Language. KQL is DuckDB SQL, running against a read-only copy of Kibu data that syncs every few minutes. Clara writes the KQL; technical users can inspect it and edit it.

Layer	What it is
Question	Plain English, asked to Clara
Query	KQL — DuckDB SQL, written by Clara, editable
Data	A read-only copy of Kibu data, synced every few minutes
Output	A table — or a chart, a summary, a CSV, a saved or scheduled report

KQL was inspired by Stripe Sigma, ShopifyQL, SOQL and PostHog’s HogQL.

Safe by default

Strictly read-only.
Permission-gated: Super Admins by default, with custom roles for everyone else.
Every piece of technology behind it, AI models included, is covered by a BAA. Kibu is HIPAA and SOC 2 compliant.

Insights is available to all Documentation customers at no extra charge.

What I built

I architected and built Insights end to end, and co-wrote the launch post with Ben Wallace, published June 24, 2026.

The platform. A multi-tenant analytics platform with a MotherDuck warehouse per organization, kept fresh by incremental syncs from PlanetScale driven by Vercel Workflows, against a 6-hour freshness SLA.
The sync engine. A sync engine built on Vercel Workflows that synced up to 16 organizations in parallel. Database credentials were short-lived, injected from Infisical at runtime, and held in Vercel's function cache for 10 minutes.
The query framework. The NL→SQL layer behind Clara, with a model-context safety gate, plus the self-service reporting UI and scheduled email delivery. Reports come back in 90 seconds or less, with five tool calls or fewer.
The move to Vercel. Migrated Insights off a standalone GCP Python service onto Vercel Workflows and cron, and brought in Infisical for per-organization secrets, which became the app's path for KMS and secrets.

Read the launch post on the Kibu blog.

What it does

Ask a question in plain English and get a table back.

Refine the table conversationally, then view it as a chart or a summary.

Export to CSV, save it, or put it on a schedule.

Hit Explain to get a plain-English summary of any Insight.

Start from a template gallery of ready-made reports — MAR, billing, documentation compliance, demographics and more.

Under the hood

Layer

What it is

Question

Plain English, asked to Clara

Query

KQL — DuckDB SQL, written by Clara, editable

Data

A read-only copy of Kibu data, synced every few minutes

Output

A table — or a chart, a summary, a CSV, a saved or scheduled report

KQL was inspired by Stripe Sigma, ShopifyQL, SOQL and PostHog’s HogQL.

What I built

I architected and built Insights end to end, and co-wrote the launch post with Ben Wallace, published June 24, 2026.

The platform. A multi-tenant analytics platform with a MotherDuck warehouse per organization, kept fresh by incremental syncs from PlanetScale driven by Vercel Workflows, against a 6-hour freshness SLA.

The sync engine. A sync engine built on Vercel Workflows that synced up to 16 organizations in parallel. Database credentials were short-lived, injected from Infisical at runtime, and held in Vercel's function cache for 10 minutes.

The query framework. The NL→SQL layer behind Clara, with a model-context safety gate, plus the self-service reporting UI and scheduled email delivery. Reports come back in 90 seconds or less, with five tool calls or fewer.

The move to Vercel. Migrated Insights off a standalone GCP Python service onto Vercel Workflows and cron, and brought in Infisical for per-organization secrets, which became the app's path for KMS and secrets.