Research and development

Future AI products we have not implemented yet.

This page shows what we are researching next. None of the four ideas below is part of our working product today.

See what we research
Connected Obsidian forms representing research, testing and deployment
R&DResearch → Prototype → Test

FUTURE R&D

What we may build next.

Every screen below is a generated concept based on current technical research—not a live product screenshot.

Clear status: not implementedResearch and concept design only. No live customer deployment exists for these four directions.
01NOT IMPLEMENTED

One dashboard for more marketplaces

Researching Amazon and WooCommerce order, listing and stock sync. Shopify is still our only live commerce connection today.

OFFICIAL RESEARCH BASISAmazon SP-API
Generated concept screen for a future multi-marketplace operations dashboard
02NOT IMPLEMENTED

AI that handles real phone calls

Researching inbound and outbound voice agents with live transcripts, language support and a safe handoff to a person.

OFFICIAL RESEARCH BASISTwilio ConversationRelay
Generated concept screen for a future AI phone-call agent with human handoff
03NOT IMPLEMENTED

AI that works inside old software

Researching browser agents for business systems that have no useful API. Every important change would pause for evidence and approval.

OFFICIAL RESEARCH BASISOpenAI computer use
Generated concept screen for a future browser agent working safely inside legacy software
04NOT IMPLEMENTED

AI agents from different systems working together

Researching MCP and A2A so independently built agents can discover tools, exchange tasks and leave one clear audit trail.

Generated concept screen for future MCP and A2A agent collaboration

WHEN RESEARCH BECOMES REAL

A concept is not a product.

We will change a direction from “not implemented” to “working” only after all four gates pass.

  1. 01

    Research

    Check the official APIs, safety limits, cost and real client need.

  2. 02

    Prototype

    Build a small version with synthetic or approved test data.

  3. 03

    Prove

    Test normal use, bad input, failure, recovery and human control.

  4. 04

    Only then launch

    Call it working only after the complete client-owned path passes acceptance.

WHAT IS WORKING TODAY?

See our real product on the Products page.