When people talk about adopting AI in business, the conversation usually turns straight to off-the-shelf tools, generic chatbots, or wrapping a single large language model in fancy branding.

At Augment, our approach is different. As part of our mission to help organisations build practical, real-world AI capabilities, we engineered a platform that solves real operational friction, protects business data and helps teams work smarter.

To prove this model, we built Kidge, a custom AI platform designed specifically for our sister company, Baltic Apprenticeships.

Kidge isn’t just a standalone chatbot. It’s a secure, enterprise-grade platform built to work directly alongside employees, tailored around how their teams actually operate.

Why we built an orchestration layer

We didn’t set out to build a proprietary large language model. Frontier AI models evolve too quickly for that to make sense. Instead, we focused on owning the orchestration, company knowledge, permissions, tools, memory and user experience around them.

Building Kidge also created a safe, controlled experimentation layer for Baltic Apprenticeships. When new models or capabilities emerge, they can be connected, tested and evaluated within the platform, without employees independently signing up for unvetted third-party products. That allows controlled, secure adoption across the business.

10 key architecture and strategy highlights

1. Platform, not just a chatbotThe chat interface is just one door. Core services are designed to be reused across future interfaces and applications, including agentic and local file workflows.
2. Right tool for the jobKidge is designed around using the simplest suitable tool for each task, whether that’s a model call, deterministic code, an existing workflow or a combination.
3. Multi-level memoryContext carries across single chats, individual users and entire multi-person project spaces, so teams don’t have to re-explain background goals repeatedly.
4. Grounded in company knowledgeKidge retrieves internal information through a custom RAG system, so users can query internal sources naturally without needing to know file paths.
5. A common tool layerKidge connects AI models to approved company systems and data sources through reusable tools and connectors, using standards such as MCP servers for portability.
6. Human-in-the-loop and governanceAs Kidge takes on more actions, people stay accountable. AI proposes actions, but execution remains controlled and respects user permissions.
7. Least privilege and scoped permissionsAI tools only access the exact information and actions needed for the immediate task, always within the scope of the user.
8. Selective privacy safeguardsSensitive information is minimised, anonymised or kept away from external models where appropriate, while safe data is processed normally.
9. Stable platform, evolving modelsKidge acts as a stable platform while the underlying models evolve, balancing capability, speed, privacy and cost.
10. Brand-aware generationKidge builds official brand guidelines and SVG logo assets directly into what it produces.

Kidge in action: real use cases

Kidge has already changed everyday work across key functional areas:

5 key lessons from building Kidge

1. Not every problem needs AISometimes deterministic code or a simple automation is faster, cheaper and more reliable.
2. The model is often the easiest partIntegrating context, connecting systems and managing permissions properly is where the real effort goes.
3. Good AI needs good company dataRetrieval quality depends entirely on how well structured and accessible your underlying information is.
4. Tools create real utilityAI becomes far more useful when it can carry out tasks through tools rather than only generating chat text.
5. Decouple the engine from the platformBuild so that today’s AI model can be swapped for tomorrow’s without breaking your workflows.

Building capabilities that last

The real value of a platform like Kidge isn’t about chasing tech trends. It’s about creating a seamless bridge between finding information, turning it into useful assets and taking action within existing business tools.

By building a secure, model-agnostic foundation, organisations can keep testing where AI genuinely adds value, give teams a safe space to innovate and remove repetitive friction.