An AI Content Team That Researches and Writes, You Keep Control

ContentMachine is the MT Labs self-hosted content engine. Each board card is an AI agent that researches keywords, drafts the article, optimizes for SEO, and publishes to your WordPress, with a Verify Before Publish gate so a human approves every post. It runs on the private AI server you rent from us, your sites and editorial history stay yours.

The problem: AI content production is fast, until it isn’t yours

Most Singapore marketing teams have tried AI writing tools by now. The first draft is fast, the next ten drafts are faster, and then something quietly goes wrong.

A post goes live with a fact you cannot back up. A keyword strategy you cannot trace. A page on your own site you would not have written yourself. The speed turned out to have a cost you only see in hindsight.

The other problem is who actually owns the content engine. Subscription content tools live on someone else’s infrastructure. The drafts, the keyword research, the editorial history, all sit in a vendor’s database. If the vendor changes their model, raises their plan, or simply goes away, your content workflow goes with it.

Most teams also discover that “AI writes the post” is the easy part. The hard part is the loop around the writing: researching keywords properly, fitting the post into your existing site structure, optimizing for SEO conventions you actually use, and getting it through a human review before it goes live. Without that loop, AI output piles up as drafts and never becomes a content engine.

The quieter pain is brand drift. A team of writers using a cloud AI tool produces five different voices across five different posts. Nothing matches the editorial line. The “fix it in editing” pass takes longer than writing the post would have.

The approach: a Kanban-driven content engine on infrastructure you rent

ContentMachine is the MT Labs self-hosted content production system. It looks like a Kanban board with a chat panel attached. You create a card for the task, an article, a topic cluster, a landing page audit, and the agent on that card streams its work into the chat. Research notes, tool calls, the draft itself, all visible as the agent runs.

The pipeline is the full content loop, not just drafting:

• Research first. The agent runs keyword research, looks at the SERP, finds gaps in your existing coverage, and writes a brief before it writes a draft. You see the brief in the chat panel and can redirect it before any words go on the page.
• Write to your voice. The draft pulls from your brand voice, your style guide, your features doc, and your existing internal link map. Those live as context files on your private server, so every article comes out shaped by your own editorial standards.
• Optimize as a workflow step. Once the draft is written, separate agents run an SEO optimizer pass, a meta description and title pass, an internal linking pass, and a keyword mapping pass. Each one is a step on the board, not a one-shot generation.
• Publish to your WordPress. The agent posts the draft directly to one of your connected WordPress sites via REST, with Yoast SEO fields populated and the post format already correct for your theme.
• Multi-site by default. A team running several brand sites can drive all of them from the same dashboard, with the right context, voice, and publishing target wired up per site.

The piece that matters most for trust is the autonomy mode. Every card has a per-card toggle between Fully Autonomous and Verify Before Publish.

In Verify Before Publish, the agent does the research, writes the draft, runs the optimizers, and then stops at the publish step and moves the card to Needs Your Input. A human reads the finished draft, edits if needed, and approves before anything goes live on the site. Fully Autonomous is available for the cases where you have already built enough trust in the pipeline. The default for new clients is Verify Before Publish, and most of our clients keep it that way.

Two operational details that quietly do a lot of work:

• Email-driven workflow. You do not have to log into the dashboard to get a draft. The agent emails you a topic prompt, you reply with the angle you want, it sends back article slants, you approve one, and the finished draft arrives in your inbox. The board is still there if you want it, but the loop runs over email for people who live in their inbox.
• Heartbeat scheduler. Idle cards do not sit unnoticed. A background process checks them and moves them to Needs Your Input or Ready based on status, so you always know which posts are waiting for you and which are done.

Because the whole engine, the agents, the context files, the publishing credentials, the editorial history, lives on the private AI server you rent from us, the editorial trail is yours. Every brief, every draft, every approved version, sits on your own infrastructure, inside Singapore. If you change your mind about how you want to write next quarter, you can edit the context files and the next post follows the change immediately. There is no vendor to ask permission from.

What it looks like in practice

A Singapore SME with three brand sites, a parent site, a B2B services site, and a consumer products site, runs all their content production from one ContentMachine dashboard on a rented private AI server.

The marketing lead creates a card on Monday morning called “topic cluster for Q3 services campaign”. The agent picks it up, runs a keyword research pass on the services site, looks at the existing posts to find what is already covered, and writes a cluster brief with a pillar article and six supporting posts. The brief lands in the chat panel within the hour and the card moves to Needs Your Input. The marketing lead reads the brief, swaps two of the supporting topics for ones they know convert better, and approves it.

Over the rest of the week, the agent works through the seven articles in sequence. Each card runs research, writes the draft, optimizes, populates internal links to the brand’s actual internal link map, and stops at the publish step. The marketing lead reviews each draft in the chat panel, makes small edits where needed, and clicks publish. The posts go live as WordPress drafts on the right site, with Yoast meta already filled in, ready for one final human read before the editor takes them public.

The week before the campaign launches, the cron scheduler runs a bulk indexing job that tells Google Search Console to re-index the new posts so they show up in search faster. The same dashboard also handles a separate research card the head of sales asked for on Thursday, a competitive landing page audit for one of their core services, with the report dropped into the chat as soon as the agent finished. None of that work touched a cloud content vendor, and the editorial history of every post is still inside the team’s own server.

A few months later, the team writes a new brand voice doc for the consumer site, because the audience has shifted slightly. They drop the new voice file into the context folder on the server, and the next post the agent writes on that site is already in the new voice. No support ticket, no vendor email, no new plan.

Where this approach has limits

A self-hosted AI content engine is not the right tool for every team.

If you publish one or two posts a year, the setup is not paid back. A simple drafting tool inside a writer’s existing workflow will be enough. ContentMachine earns its keep when there is a real content cadence, several sites or a serious topic cluster strategy, and a real need to keep the editorial history and the publishing credentials on your own infrastructure.

AI content production is not a substitute for editorial judgement. The Verify Before Publish gate exists because a draft, however well-researched, is still a draft. We will not pretend otherwise.

The cases where teams have run into trouble are the cases where they removed the human review step to ship faster, then published something they would not have signed off on. The dashboard makes the loop fast enough that the human read is the smallest cost in the workflow, and we recommend keeping it.

The other honest limit is original reporting. ContentMachine is excellent at the production layer, research, drafting, SEO, internal linking, multi-site publishing, but it does not do journalism. If your content depends on interviews, on-site visits, or proprietary first-party research, the agent will help you turn those raw materials into a polished post, but it will not generate the materials themselves. We will tell you which side of that line a given content brief sits on.

For how a ContentMachine setup would be sized to your team and your publishing cadence, drop us an email for more information rather than a generic figure.

MT Labs helps companies across Singapore deploy AI tools they actually own. Private infrastructure, no recurring cloud subscriptions, and a setup built around how your team already works. AI isn’t right for every workflow, and part of our job is telling you where it isn’t. Get in touch and we’ll walk through where it makes sense, and where it doesn’t, for your business.