Tray.io Unveils Merlin Agent Builder for AI Data Integration

The glittering promise of Artificial Intelligence often overshadows the gritty, unglamorous reality of its sustenance: data.

While the tech industry frequently reminds us of the “garbage in, garbage out” adage, the truth is often more nuanced, and arguably, more frustrating.

It’s less about deliberately feeding AI poor information and more about the challenge of working with good ingredients, but from a chaotic, unorganised “garbage heap” where vital components are scattered, obscured, and often lost.

This messy reality, far from the pristine laboratories envisioned, is where the true limitations of AI often emerge.

Imagine an AI system, hailed for its intelligence, yet repeatedly losing context, offering incomplete answers, or forcing users to re-explain themselves in every interaction.

This isn’t a failure of the algorithm itself, but a symptom of poorly aligned data channels, fragmented knowledge bases, and a critical lack of fluid “information transports”.

It’s a pain point acutely felt by both the end-user, who experiences clunky and disconnected AI “agent” interactions, and the IT and AI teams struggling behind the scenes to make these systems actually work.

The core issue lies in the bedrock of AI: data integration.

AI agents, designed to automate tasks and provide intelligent assistance, are only as effective as the data they can access, understand, and retain.

Without a robust, intelligent data fabric, these agents operate with narrow knowledge, suffer from memory limitations, and struggle to align the right Large Language Models (LLMs) to specific use cases, especially in complex, multi-agent environments.

This is the chasm that Tray.io, an AI integration and automation platform, aims to bridge with its new Merlin Agent Builder.

Rich Waldron, co-founder and CEO of Tray, likens Merlin to a sat-nav system for AI agents, ensuring they don’t just get built, but are actually driven and used effectively.

“This is because agents lose context, rely on narrow knowledge and force users to start from scratch in every session,” Waldron observes, highlighting the user-side frustration.

For developers, the struggle is equally profound: “IT and AI teams struggle to align the right LLMs to the right use cases… Without flexible deployment options, it’s hard to meet users where they work.”

The solution, he insists, lies in agents needing “smarter data access, built-in memory, LLM flexibility and tailored user interactions” – all designed for sustained usage, not just fleeting prototypes.

Alistair Russell, Tray’s co-founder and CTO, clarifies that Merlin Agent Builder isn’t merely a “services wrapper” – an intermediary abstraction layer.

Instead, it’s a “fundamental part of our product,” built for ease of use and scalability, working at a deeper, more foundational level where data is born.

“Our platform handles chunking [diving data into more manageable, digestible pieces] and embedding at the source, ensuring each data source is optimally segmented and vectorized so agents are grounded in high-signal, relevant context,” Russell explains.

This meticulous data preparation, he argues, leads to “fewer retrieval failures, more reliable decisions and agents that reason and take action.”

Consider the common scenario of an IT help desk agent tasked with automating ticket resolution.

The knowledge it needs – past tickets, solution articles, internal policies – is invariably scattered across disparate, siloed systems.

When the agent cannot find the precise, relevant data, user trust erodes, and conversations quickly fall flat.

Russell laments the current state where such agents forget prior context between conversations due to token limits and storage constraints, requiring custom engineering for long-term memory.

Tray’s Merlin Agent Builder tackles this directly, offering “maximum short- and long-term memory” so agents can automatically track session history and refer back to previous interactions, eliminating the frustrating cycle of repetition.

Tray’s capabilities firmly place it within the Integration Platform-as-a-Service (iPaaS) market, a domain acknowledged by analyst house Gartner, which features Tray in its Magic Quadrant for iPaaS.

This market is a vibrant tapestry of specialists, from giants like Oracle with its Integration Cloud and hyperscalers like AWS, Microsoft, and IBM, to more focused players.

Workato, for instance, is known for its AI-fuelled automation and low-code tooling.

Boomi, often dubbed the “Windows of the iPaaS market” for its user-friendliness and broad coverage across cloud, hybrid, and legacy systems, offers a different flavour.

MuleSoft, now part of Salesforce, is staunchly API-first, while Jitterbit and Zapier cater to smaller businesses seeking point-and-click simplicity.

SnapLogic attracts users with its speedy deployment and Iris AI service, though its pricing structure can deter some.

Niche players like Celigo provide specialised integration for sectors like e-commerce, emphasising governance and security.

As the digital landscape evolves, AI will undoubtedly become an even more prominent purchasing factor within the iPaaS market.

This will drive iPaaS technologies beyond traditional data centers, pushing them towards the smart edge in the Internet of Things (IoT), making real-time data streaming an indispensable feature.

While headlines often celebrate the dazzling new functions of “agentic AI”—whether for HR, military-grade software, or cloud-native Kubernetes orchestration—the true unsung heroes are the advancements in data configuration and integration.

These are the less glamorous but utterly crucial innovations that enable smart routing, ensure continuity across complex multi-turn interactions, and prevent AI from drowning in its own data “garbage heap.”

It’s about giving agents the right data direction, guiding them – and us – towards genuine utility and seamless interaction.