Skip to content

SaaS and platforms

Your product, with Aira inside

A B2B SaaS platform gives every customer an analyst inside its own app. The platform's backend mints a short-lived scoped token carrying that end user's row filters; Insighter compiles those filters into every query as mandatory conditions, so neither the caller nor the model can widen them. A drop-in panel renders in the platform's own brand, and usage is metered per tenant so they can rebill.

The week this replaces

Every tenant wants to ask questions of their own data, and only their own data.

Row-level security has to hold against the model itself, not just against the UI.

Rebilling AI usage per customer fails the day the cost cannot be attributed to one tenant.

POST /v1/threads/:id/messages { "message": "Summarise this account's usage risk" }

The stream returns text deltas for typing effects, a chart event with the structured spec (send it to /v1/render/chart for a presentation-ready SVG), and a files list with the CSV Aira produced. One API, whole answer.

Text
Charts
Files
Voice
Illustrative conversation. On your data every answer carries sources you can open, and every query is read-only.
  1. AnsweredText, a chart spec and the produced files, on one streaming response
  2. ActedRow filters compiled into every query as mandatory conditions
  3. HeldA token cannot widen its own scope, and a denied metric never appears

Metered per tenant, so you can rebill it without reconciling anything.

And then it keeps running

Answer inside the row filters compiled into the token, return text, chart spec and files on one stream, and attribute the cost to the calling tenant.

On every API call from your platformCeiling: set per token
Connects toREST APIYour app
Doing the workScoped tokens, 15 minute ceilingRow filters enforced in the compilerDenied metrics stay invisiblePer-tenant credit attributionDrop-in panel or REST

Where this goes

Governed autonomy

Work that runs without you, inside limits that only ever narrow, with a record of what it did, what it refused, and what it would not do without asking. Everything on this page is one step toward that.