LEISSONCREATIVE
All work · Flagship

Maris — a domain-neutral assistant kernel that carries two products

A kernel extracted from Mati (@leisson-labs/*) that runs a WhatsApp assistant with text, voice, images and documents; the same kernel will carry the iOS app and Operations OS.

Maris — a domain-neutral assistant kernel that carries two products — desktop
maris-web.vercel.app2026-09-13
Open the product ↗
StatusRunning · WhatsApp · iOS lab
Period2026-08 →
StackTypeScript · @leisson-labs/* · Railway · Meta Cloud API · OpenAI · iOS
What this work shows a buyer
Type of evidenceStudio's own product. The studio's own product: the assistant core and the WhatsApp product.
Whose problem it illustratesA workflow that currently runs by hand in messages: text, voice, images and documents.
What it does not proveThe public introduction page does not prove that the access-restricted product works in full; the WhatsApp product is not publicly available.
What you can order from thisAI workflow pilot · from 990 €
maris · evidence
3 claims · 0 with a public source, 3 with a private source (shown on request)
4 input modalities: text, voice (STT+TTS), image, PDF/text → source: leisson-labs/maris · private, shown on request
16 E1 slices (E1.2–E1.17) shipped as PRs in one night → source: leisson-labs/maris PRs · private, shown on request
0 coupling to AgroNutikas repos — clean separation → source: leisson-labs/assistant-kernel · private, shown on request
0/3 publicly sourced

Problem

Mati — the AgroNutikas WhatsApp assistant — got good enough that two other products wanted the same core. Copying would have meant three codebases drifting apart. What was needed was a kernel that knows nothing about agriculture.

System

On 2026-08-28 the decision was variant B: kernel extraction. A new GitHub organisation leisson-labs, a private repo assistant-kernel, zero coupling to AgroNutikas repos. Mati's domain-neutral core was ported into @leisson-labs/* packages. The first consumer is Maris: a WhatsApp product on Railway with its own Meta app, supporting text, voice (speech-to-text and text-to-speech), images (vision) and PDF/text document reading, with a durable mailbox. The second consumer is the iOS app (maris-app), the third is Operations OS phase 9. The AgroNutikas platform adopts the kernel later through a parity gate — not before the kernel does everything Mati does today.

Decisions

Kernel before product: interfaces are contracts, not conventions. Every slice has its own worktree, PR and CI proof; the skills maris-app-slice-runner, maris-strukturne-varav and maris-kanoni-peegel keep document and code from drifting. Three secrets (WhatsApp token/secret, OpenAI key) stayed with a human; the system never prints them.

Metrics

E1.2–E1.17 — 16 slices — were built in one night on 2026-08-29 (source: leisson-labs/maris PRs). Four input modalities in operation. There is no public URL — Maris is a private Meta app, which is why the status is “Running”, not “Live”.

Stack

TypeScript monorepo @leisson-labs/*, Railway (WhatsApp), Meta Cloud API, OpenAI (STT/TTS/vision), iOS (maris-app), maris-web on Vercel.

Next step

One of your manual workflows, on the same principle

One manual workflow needs a controlled automation trial.

Get started: AI workflow pilot · from 990 €