AX · AI adoption, outsourced

Fast AI adoption,
from day one.

Services AI can use from day one, at half the cost, that anyone can take over.

The team that built Akan.js develops on Akan.js 3. The moment we deliver, Claude Code, ChatGPT and an in-page AI agent can work in your service.

Talk AI adoption Our three promises

AI built in
Half the cost, twice the speed
Code anyone can take over

Turn AI onone line of code apart

Inquiry.Util.tsx

export const Reply = () => {
+ const send = st.tool("sendReply", { confirm: true }).exec(st.do.sendReply);  return <Button onClick={st.do.sendReplysend}>
};

shop-admin.app/inboxMCP OFF

RefundI'd like a refund

ShippingWhen will it arrive?

ExchangeCan I swap the size?

I'd like a refund

Write a reply

Send replysendReply

✦Agent

Draft replies to the refund requests

Drafted, waiting for your approval

Fast AI adoption✦AX✦Vibe coding✦AI agents✦MCP server✦Automation✦AI support✦In-house AI assistant✦Fast AI adoption✦AX✦Vibe coding✦AI agents✦MCP server✦Automation✦AI support✦In-house AI assistant✦Fast AI adoption✦AX✦Vibe coding✦AI agents✦MCP server✦Automation✦AI support✦In-house AI assistant✦Fast AI adoption✦AX✦Vibe coding✦AI agents✦MCP server✦Automation✦AI support✦In-house AI assistant✦

Claude✦ChatGPT✦LLM integration✦Approval flows✦Card OCR✦AI writing✦AX training✦Claude✦ChatGPT✦LLM integration✦Approval flows✦Card OCR✦AI writing✦AX training✦Claude✦ChatGPT✦LLM integration✦Approval flows✦Card OCR✦AI writing✦AX training✦Claude✦ChatGPT✦LLM integration✦Approval flows✦Card OCR✦AI writing✦AX training✦

Three promises

Three worries about outsourcing,
answered one by one.

“Adding AI means another project, right?”

AI comes built in

Claude Code, ChatGPT and an in-page agent work in it the moment it ships. “Please add AI” never becomes a new quote.

“Web and apps separately cost double.”

Half the cost, twice the speed

One codebase covers web, mobile and desktop. More platforms don't multiply the bill.

“When the agency left, nobody could fix it.”

Code anyone can take over

Lint enforces one set of rules, so your team, another agency or a coding agent can pick it up. Non-developers fix things too, through tickets and vibe coding.

01AI comes built in

Adding AI never
becomes a new quote.

The usual AI adoption builds a chatbot, an agent-only API, tool schemas and a permission model, then re-syncs them on every change. On Akan.js 3 the screens and server people use are the agent's tools, so that whole bill disappears.

The usual AI quoteNo. ????

Chatbot UIextra
Agent-only APIextra
Tool schemasextra
Permission modelextra
MCP serverextra
Re-sync on every changeforever

Total6 extra builds

Akansoft AI quote

In-page agentbuilt in
Agent tools1 line each
MCP serverbuilt in
Permissionsexisting guards
Approval cardsbuilt in
AI modelyour pick

TotalOne line

Four kinds of AI in what we deliver

Customer screens, the back office, outside AI and everyday workflows. Turn on the ones you need.

In-page agent

AI agent for customers

Customers say it and the agent works the screen: rebooking, order lookups, filling forms. It reads photos and documents and listens too.

Cancel my order and add it again in M

cancelOrder · awaiting approvaladdToCart

Back-office agent

AI agent for your team

“Sort the open tickets and assign owners” is the whole job. Agents take the repetitive work in your back office.

Sort open tickets and assign owners

assignInquiry × 12Assigned to 4 people.

MCP server

Live in Claude Code and ChatGPT the day it ships

No integration project to wait for. Register one address on delivery day and customers use your service in ChatGPT while your team works it from Claude Code. The AI acts only with that user's permissions.

Claude CodeConnected

$ claude mcp add --transport http shop https://your.app/mcp

ChatGPTAdd connectorhttps://your.app/mcpConnected

Delivery day, one line of setup

AI features

AI inside the workflow

Card photos become contacts, product info becomes posts, comments become reply drafts. AI lands inside the screens you already use.

Card photoContactProduct infoMarketing postCommentReply draft

Fast, and on a leash

The question we hear most is “what if the AI does something wrong?” So the guardrails are in the framework, not an add-on.

Same rules as people

AI passes the very guards people pass.

Approval cards

Anything that changes data waits for a person to approve.

Human-only acts

Acts like refunds that a person must do never reach the AI's list.

Secrets stay put

Hidden and secret fields are stripped before they reach a model.

Revoke any time

Revoke a connection and the very next call is refused.

Rate limits

MCP calls are capped at 120 a minute and 8 at once, per user.

Pick your model

OpenAIAnthropicDeepSeekGroqOpenRouterOllamaOllama can run on your own servers.

02Half the cost, twice the speed

More platforms,
not more cost.

Separate web, mobile and desktop code multiplies both building and upkeep. On Akan.js 3 one codebase ships to six platforms.

1 codebase

Web
iOS
Android
macOS
Windows
Linux

Real footage

One stay-booking service running on iOS, macOS, web and Android at once. Tap on Android and every other screen follows. Windows and Linux desktop apps share the same code.

The usual agency

One feature,
fixed three times.

Web code · fix
App code · fix
Desktop code · fix

Separate web, mobile and desktop code, so upkeep grows with every codebase.

Akansoft

One codebase
runs every platform.

1 codebase · fix

Web
iOS
Android
macOS
Windows
Linux

Fix it once and web, mobile and desktop all change together. More platforms, still one codebase to manage.

×8

One line, eight places

Add a field and the database, API, state and screens change together. Nothing to chase.

Pre-built blocks

Sign-in, uploads, admin, chat, boards and notifications are there from the start.

Coding agents that know the rules

Coding agents plan, apply and validate through a workflow. People review the business change.

−32%

Start small, grow big

Go from one server to a Postgres cluster without a rewrite. 32% less memory than v2 trims the server bill.

03Code anyone can take over

Change vendors,
keep the service running.

The scariest part of outsourcing is after the builder leaves. Akan.js 3 turns how code is written into rules that lint enforces, so it reads like one author wrote it.

Store for company A

lib/order/

order.constant.ts

order.document.ts

order.service.ts

order.signal.ts

order.store.ts

Order.View.tsx

Booking app for company B

lib/booking/

booking.constant.ts

booking.document.ts

booking.service.ts

booking.signal.ts

booking.store.ts

Booking.View.tsx

Open any project: same places, same names.

Your own developers

Know the rules and you know which file holds what.

Another agency

A documented open-source framework reads on first sight.

Coding agents

AGENTS.md and the workflow MCP keep Claude Code or Codex inside the rules.

In every handover

AGENTS.md rulebookLint that enforces the rulesTraceable logs per requestBuilt-in backup and restore

No developer? You still fix the service

AX doesn't end at adding AI. After delivery, the people doing the work should be able to change the service themselves. Akansoft paves that road with Akan Office and Akan Cloud.

“Build it well, and later we can maintain it by vibe coding, right?”

Yes. That's how we build it from day one.

Akan Office

One ticket and you're in the build

Leave what needs fixing as a ticket in Akan Office. Claude Code reads it through the Akan Office MCP, makes the fix, comments the result and moves it to review. Planners and operators join the build from the same board.

#128 · Ops team

Change the payment-done message to “Order received”

Claude CodecommentTicketreviewTicket

Fixed. Sent to review.

Akan Cloud

Fix it yourself by vibe coding

Nothing to install: ask Akan Cloud for the change in the browser, check the preview and deploy in one click. The code lands in your own GitHub repository and follows the same rules as what we delivered.

cloud.akanjs.com

Add a winter sale line to the main banner

Preview is ready.Deploy

AX training included

Every AX engagement trains your people. Your team learns to change the service themselves with vibe coding and Claude Code.

04Proof

We did it first,
on Akan Office.

Akan Office, the back office we run client work on, is already through AX. Tickets go onto the board from a chat message, and quotes and tickets are handled from Claude.

325

Tools open over MCP

Page prompts

Built-in AI features

Backlogs 11

Sign-in

Sign-up

Reset password

Edit profile

Payments

Order history

Push

Admin stats

Search filters

Reviews

Support chat

✦Office agent

Add the 11 backlog items for the next sprint

addBacklogTicket × 11

Added all 11. Check the board.

Projects delivered

In maintenance

In development

12,319

Commits

05How adoption runs

AI adoption
in four steps.

STEP 01
Diagnose
We pick, together, which jobs go to agents, for customers and for your team.
STEP 02
Transform
We move your service onto Akan.js 3 or build it new. The screens and server become the agent's tools.
STEP 03
Open to agents
Turn on the in-page agent, MCP access and AI features, and set the approval rules.
STEP 04
Run and train
We maintain it with progress shared as Akan Office tickets, while your team learns to make changes with vibe coding and Claude Code.

06FAQ

FAQ

How long does AI adoption take?

It depends on scope. Since agents aren't built separately, the work AI adoption usually adds simply isn't there. We set the schedule together.

Does my service have to be on Akan.js already?

Yes, though it takes remodeling your service onto Akan.js 3, and AI comes along in the same move. We scope that remodel and its cost openly.

Can we maintain it ourselves after the project?

The code follows one set of rules and ships with an AGENTS.md rulebook, so your team, another agency or a coding agent can take over. Without a developer, leave a ticket in Akan Office or vibe-code the change in Akan Cloud, and the AX engagement teaches your team how. We agree the handover scope in the contract.

Which AI models can we use?

OpenAI and Anthropic, plus OpenAI-compatible hosts like DeepSeek, Groq and OpenRouter, and Ollama on your own servers.

What if the AI does something wrong?

It passes the same rules as people, anything that changes data stops at an approval card, and human-only acts never reach its list.

Does customer data go straight to the AI?

Hidden and secret fields are stripped first, and the in-page relay keeps no sessions or conversations. The model can live on your own servers.

What does it take to use it from Claude or ChatGPT?

Add the service address to the MCP client and sign in. OAuth 2.1 grants one user's permissions, revocable any time.

07Talk AI adoption

Start your
AI adoption here.

Write the job you'd hand to AI in one line. We'll work out where to start together.

Prefer email? contact@akanjs.com

Fast AI adoption,from day one.

Three worries about outsourcing,answered one by one.

Adding AI neverbecomes a new quote.

Four kinds of AI in what we deliver

Fast, and on a leash

More platforms,not more cost.

Change vendors,keep the service running.