Marketing Automation / AI Agents / SaaS

JK Automation OS — Instagram, Facebook and WhatsApp Automation Platform

Our own multi-tenant SaaS: content, scheduling, comments, DMs and Meta ads with an approval gate, plus a WhatsApp AI agent that reads invoices and works inside a client's ERP.

Built by the JK Software Solutions engineering team.

JK Software Solutions View Demo
JK Automation OS — Instagram, Facebook and WhatsApp Automation Platform

The Challenge

A business that sells on Instagram, Facebook and WhatsApp ends up running three apps, a designer, a spreadsheet of posts and a phone full of unanswered messages. Generic tools publish or reply on their own, which is risky for prices and claims, and none of them reach the company's own system: a manager still types invoices and checks project numbers by hand. We wanted one platform that handles the channel work, lets AI do the drafting and the reading, and never acts on anything important without a person approving it.

Our Solution

Instagram and Facebook: the content studio turns a brand's products, projects and approved claims into posts, carousels and Reels. Every draft is checked by a brand guard, and anything with prices, discounts or regulated claims always goes to a human. The approval is bound to the exact revision and its content hash, so an edit after approval sends it back to review. Approved posts go to a calendar and the scheduler publishes them on time through the official APIs. Comments and DMs arrive by webhook into one inbox, with suggested replies and the publishing history and per-post metrics next to them. Meta ads: a campaign is prepared as a draft with its audience, budget and a maximum spend. Pushing it to Meta creates the campaign, ad set, creative and ad as PAUSED objects only. Activating it is a separate human step, and the approval request records the spend cap and the content hash. WhatsApp AI agent: the platform works as a WhatsApp Cloud API Tech Provider. A company connects its number, and each employee links their own WhatsApp by sending a one-time code from their signed-in account. A phone number alone never grants a role, and each linked number gets its own areas: one person sees projects, another finance. Customers only get the knowledge base and a request for a human. Staff ask in plain Arabic and the agent answers through a fixed set of tools with schemas, never SQL or open API access. A manager can send a photo or PDF of an invoice: Claude extracts it in a call with no tools, the server recomputes quantity times price, line totals and subtotal plus VAT, flags a number that was already recorded, and lets the manager split lines between projects. Nothing is written to the client's ERP until the manager approves the summary on WhatsApp; high-risk approvals can require a PIN. Voice notes are transcribed with whisper.cpp in a separate worker. The agent sends Excel and PDF reports as documents, and follow-ups and reminders go out on schedule. Platform: organizations, roles (owner, admin, employee), invitations and an audit log. Billing runs on credits with plans and limits; the AI gateway meters every request by tokens and cost and routes customer chats to Haiku, staff and document work to Sonnet, and creative writing to Fable. Ad Studio writes a script, voices it and renders a vertical video on Remotion or HyperFrames. Render jobs go to a worker on the server or to a remote GPU worker, and every final is mastered with FFmpeg and scored by an automated review. Client companies get a portal with their own logo, credits, usage and linked numbers. The interface works in Arabic and English.

Objectives

Plan, design, approve and publish Instagram and Facebook content from one workspace. Answer comments and DMs from a shared inbox, with AI drafts the team can edit. Prepare Meta ad campaigns with a spend cap, created paused until someone approves them. Give companies a WhatsApp agent that answers customers and serves managers with real company data. Read invoices from images and PDFs, check the arithmetic and VAT, and record them in the client's ERP only after approval. Isolate every organization, control access by role and keep an audit log of every action. Charge in credits and route each AI task to the cheapest model that does it well.

Outcomes

JK Automation OS has been in production at marketing.jksoftwaresolutions.com since October 2026, and JK Software Solutions uses it for its own Instagram and Facebook accounts. Since November 2026 the WhatsApp agent has served a contracting company in Riyadh through its ERP. In live tests from a real phone, a number was linked with a code, the agent answered questions from the company's data, produced a payroll Excel report, read four invoices sent as photos and PDFs, and transcribed a voice note. Meta's review of the Instagram and Facebook permissions is still in progress. The codebase has 58 server modules, 95 database tables, more than 400 API routes and about 700 automated tests, built in more than 230 commits. Client names and data are not shown: the screens use JK's own organization or demonstration data.

Challenges

The hard rule was that AI may draft and read but never act alone. Every write goes through an approval bound to an operation id, a hash of the tool input, the target's version and a single-use nonce, so a stale or altered approval is rejected. Text inside an invoice image is treated as untrusted data: turns that carry a file only get read-only tools, so nothing written on a document can trigger a write. Webhooks from Meta are verified against the X-Hub-Signature-256 signature and stored once per message, so retries never create a second action. Each organization's data is scoped on the server, and tokens are stored encrypted. Running whisper inside the API process ran out of memory on the first real voice note, so transcription moved to the render worker's own lane. AI cost has to stay below the price of a credit, so each request is routed by the kind of task: a customer reply on Haiku costs a fraction of a cent.

Our Role

JK Software Solutions designed, built and runs the whole product: the Fastify server split into business modules, the React interface, the PostgreSQL model through Drizzle, the Meta and WhatsApp integrations, the AI gateway and agent, billing, the video engines and render workers, and the production deployment with release folders and rollback.

Technologies

TypeScriptFastifyReactPostgreSQLDrizzle ORMClaude APIInstagram APIFacebook Graph APIMeta Marketing APIWhatsApp Cloud APIRemotionHyperFramesFFmpegwhisper.cppVitest

Related Service

AI & Automation Solutions

Project Documentation

System Media Gallery

9 documented assets

Web Application

3

Administration

3

Project Media

3

Have an idea but not the technical details?

Send us the idea and we will respond within 24 hours with the next practical step.

The initial consultation is free with no commitment, and for suitable projects it can include discussing an early visual concept.

JK Automation OS — Social and WhatsApp Automation