AiGen-One — an enterprise LLM platform that turns AI from personal tools into company workflow
Every team has its own AI now. ChatGPT here, Gemini there, a personal agent running in someone's browser. Individually, people got faster. The company, somehow, didn't.
Look closely at where the time goes and it's rarely the AI call itself. It's everything around it: results copied by hand into the next tool, files moved between systems, and someone deciding — informally, over the shoulder — who's allowed to feed which data into which model.
Public LLM services add a second wall. The documents that would make AI genuinely useful — contracts, internal specs, customer records — are exactly the ones you can't paste into a public chat window.
The AI works. The workflow doesn't. That's the gap this platform was built to close.
We build systems for clients — that's the business. And in the AI era, it's become clear that "we can build it" is no longer the differentiator it used to be. Generated code made building cheap. What stays scarce is what surrounds the build.
What a company actually needs is a place where AI capability accumulates — business systems, data connections, permissions, operating knowledge — compounding instead of evaporating with each project. We decided to build that place ourselves: in-house, with our engineering team in Ho Chi Minh City. Not a demo for a pitch deck — a platform that runs real plugins today, built the same way we build for clients: Git, reviews, migrations, audit.
AiGen-One started as a business-app generator. It evolved into something broader: three kinds of capability on one base — business systems developed from scratch, AI skills that can be reused across teams, and agents that run multi-step workflows with people in the loop.
Under it all sits one control plane with four layers: the Portal people enter through, the Projects that hold shared context, the execution layer of Skills, Agents and Plugins, and the Custom Apps layer where databases, APIs and production systems live. Most of the design effort went into the connections between those layers — because that's where enterprise AI actually breaks.
Honestly, the models were the easy part. Nearly everything that made this hard sits around them: who is allowed to connect what, how a change reaches production, and what you can prove about it afterwards.
The first screen isn't an AI feature. It's the company's work: announcements from admins, internal notices, today's schedule, pending approvals, recently used tools. AI chat sits inside that screen. If people have to remember to open a separate AI tool, they won't — so the AI lives where the workday already starts.
Every company has one person whose prompts are unreasonably good, and that know-how usually dies in a chat log. On AiGen-One, a prompt is packaged into a Skill: what it does, the instruction set, a verification run, example calls. "Explaining good usage" becomes "shipping a callable function." Broader capabilities ship as plugins into the business menu — seven run on the platform today, from proposal drafting and spec generation to a Gmail inquiry handler and multilingual chat.
An agent chains Skills, plugins, APIs, MCP connections and database operations into an actual work sequence. The design rule is simple: any step can demand a human confirmation before it executes. Automation that can't pause for judgment isn't something you can safely hand a workflow to.
Connections — API, MCP, OAuth, Git, database — are managed in one place instead of scattered across personal tokens. Changes pass AI review, engineer review and migration control before they reach production, and every action lands in an audit trail. Systems developed from scratch deploy through the same flow: generated code, Git, pull request, migration, release. One pipeline, whether a human or an AI wrote the first draft.
No prompt training, no separate AI tool. Staff open the portal they already use for announcements and approvals, and call Skills or Agents from chat — draft this proposal, summarize this thread, run this check. The expertise is packaged inside the Skill, so the quality doesn't depend on who's asking.
Managers see usage, permissions and projects in one place: which teams use which capabilities, what's waiting for approval, what's actually getting used. Good patterns get promoted deliberately — pushed into the business menu for everyone — instead of spreading by rumor.
New models, new integrations, scratch-built systems, operational fixes — they all land on the same platform instead of becoming one more disconnected tool. Code ships through Git, pull requests, migration control and audit, whether a human or an AI wrote the first draft. For an engineering organization, that's the difference between projects that evaporate and capability that compounds.
| Before | After | ||
|---|---|---|---|
| Where AI lives | Personal accounts and browser tabs | → | The company portal, where the workday starts |
| Good prompts | One person's know-how, buried in chat logs | → | Skills — packaged, tested, callable by anyone |
| Multi-step work | Copy-paste between tools, by hand | → | Agents run the sequence; humans approve the gates |
| Control & audit | "Who used what?" had no real answer | → | Permissions, reviews, and a full audit trail |
AiGen-One runs at Linnoedge today — seven working plugins, from proposal drafting to multilingual chat. It has reached the stage where a company can start a PoC on it: the entrance, the skills, the agents and the governance are live. That's the honest place it's at.
Linnoedge
Platform design, build & operation — Ho Chi Minh City
Keep the model access inside your own environment and control what leaves it. In AiGen-One, documents stay in the company's knowledge base, retrieval runs against your own data (RAG), and every external connection — including the LLM itself — goes through managed credentials, permissions and audit logs. Employees work through the company portal, not personal accounts, so nobody has to paste internal documents into a public chat window.
Four layers. An entrance — a portal where announcements, tasks and AI chat live together. A context layer — projects holding documents, knowledge and issues. An execution layer — skills, agents and plugins that do the work. And the company-systems layer — databases, APIs, permissions, audit, production operations. The value is in keeping the four connected; most stalled AI programs have only the execution layer.
RAG first, in almost every case. When we designed AiGen-One we chose retrieval over fine-tuning for company knowledge: it stands up in about a week, the source documents stay inspectable, and updating knowledge doesn't require retraining anything. Fine-tuning becomes worth discussing when you have one narrow, high-volume task with stable patterns — not as step one.
Design the checkpoints in — don't rely on the model behaving. AiGen-One agents are built from explicit steps, and any step can require a human confirmation before it runs. System changes go through the same gates as human-written code: engineer review, Git, migration control. Every action lands in an audit trail, because "the AI did something we can't trace" is the failure mode the whole governance layer exists to prevent.
A private RAG foundation on your own documents takes roughly a week. From there, the honest answer is: it depends on the first workflow you want to run, because that part is real engineering — a business app, its data connections, its approval steps. That's why the rollout is staged: entrance first, one workflow second, reuse third, company-wide standards last.
No — and that's a design decision, not a hope. Good prompts are packaged as Skills: named functions with a defined purpose, instructions and tested examples that anyone can call from chat. The person who writes a great prompt turns it into a company asset once, instead of explaining it in meetings forever.
No. If what you need is one chatbot on your website, a platform is overkill — a scoped integration is cheaper and faster. This approach fits companies that already have several AI use cases, real data-governance constraints, and internal systems the AI needs to touch. If that's not where you are yet, start smaller. The platform question comes later, and it keeps.
Thinking about an enterprise AI platform?
30 minutes. No cost. Comes with a written next-step summary.