Sentient Pushes Open AI Infrastructure into the Global Spotlight

Sentient AI’s launch of Recursive Open Meta-Agent (ROMA), in September 2025 marks a pivotal moment in the evolution of open-source AI frameworks in India. Far from being “just another model,” ROMA represents a new way of thinking about how AI systems should reason, coordinate, and solve complex problems. 

Built as a hierarchical meta-agent framework, ROMA abandons the monolithic design of traditional large models and instead orchestrates a structured tree of specialised agents, each responsible for solving part of a larger task.

The result? A transparent, auditable, high-performance system that challenges the dominance of closed AI architectures and sets a new direction for the future of decentralised intelligence.

What ROMA Actually Is, and Why It Matters

As Himanshu Tyagi, co-founder of Sentient, told AIM, “ROMA isn’t a model; it’s a reasoning architecture. That distinction is crucial as models generate outputs, while reasoning architectures determine how those models think, plan, and coordinate to solve problems.”

ROMA’s defining innovation lies in its recursive hierarchical task tree. At its foundation is an “Atomiser” that decides whether a task is atomic, directly executable, or requires further planning. If planning is required, control moves to the “Planner,” which decomposes the goal into subtasks, each of which is recursively fed back into the Atomiser. This allows ROMA to scale to long-horizon, multi-step reasoning tasks that typically confound single-model systems.

This architecture is not only algorithmically efficient but also involves minimal complexity structurally. As Tyagi said, “The algorithmic innovation is elegant yet simple (simple recursion), leading to phenomenal performance improvements.”

By treating reasoning as a planning problem rather than a generation problem, ROMA reduces compounding errors, improves long-term coherence, and enables parallel execution, allowing independent subtasks to be solved simultaneously, the founders said. 

Tyagi argued that the narrative around openness and safety, often used by Big Tech to justify closed systems, is fundamentally flawed. “The argument that openness inherently risks safety is a form of semantic gymnastics used to justify building opaque, black-box models that monopolise control and knowledge,” he added. 

Instead, transparency is woven directly into ROMA’s architecture. Because tasks propagate through an explicit recursive tree, every stage is traceable, auditable, and open to inspection. Builders can review how tasks were broken down, how context flowed, how tools were invoked, and where failures happened, something impossible in monolithic models.

ROMA is fully open-source and available for inspection and modification, reinforcing Sentient’s belief that the future of AI must be democratic, not centralised.

The Global, Not Local, Roots of ROMA

Although India features prominently in Sentient’s leadership and mission, ROMA’s design philosophy is not geographically bound. “Our team, researchers and vision are fundamentally global… AI should be borderless and not controlled by any single entity or sovereign state,” Tyagi clarified. 

This borderless ethos directly shaped ROMA’s modular architecture. Contributors can plug in models, tools, datasets, or agents without relying on proprietary backends or closed APIs. More importantly, Sentient’s ecosystem is designed to reward contributors, not just extract value from them.

As Tyagi put it, ROMA’s architecture is meant to “provide the necessary architectural openness while also offering a transparent, traceable meta-agent framework that allows this global talent pool to build and control their own sovereign AI models.”

In a world where AI infrastructure threatens to be monopolised by a handful of technology giants, the open-source model offers not just an alternative, but a counterforce.

Why Developers Are Paying Attention

ROMA represents the next stage in a broader mission for Sentient co-founder Sandeep Nailwal, who previously helped with Polygon’s decentralisation of finance.

“With Sentient, we’re doing the same for intelligence. We want to stop AI from becoming a gated, private resource controlled by a handful of companies. Intelligence should be a public good,” he told AIM. 

Just as Web3 democratised access to financial infrastructure, ROMA aims to democratise access to cognitive infrastructure. If Polygon built the rails for decentralised value, Sentient seeks to make the rails for decentralised intelligence.

Nailwal views this as critical for global equity. “If it’s locked away, it kills participation… From startups to sovereigns, everyone needs access to intelligence as a public utility,” he said.

The company’s $1.2 billion valuation, as Nailwal explained, is anchored in investor confidence that open-source innovation scales faster and more efficiently than closed systems. 

“No single company, no matter how large, can out-innovate a global network of open builders moving in parallel,” he clarified.

A key component of this ecosystem is GRID, Sentient’s open intelligence marketplace, where developers can contribute tools, models, and agents and earn based on their usage. This solves one of the fundamental challenges of open source: sustainability.

More importantly, the open source framework offers a philosophical and structural alternative to the consolidation of AI power. “Sentient represents Web3 for intelligence… We’re decentralising AI itself,” Nailwal summarised. 

The startup said in its blog that ROMA’s prototype search agent, dubbed “ROMA Search”, achieved 45.6% accuracy on the SEALQA subset known as “Seal-0,” which tests complex multi-source reasoning, outperforming the previous best system, Kimi Researcher, at 36% and the proprietary Gemini 2.5 Pro at 19.8%.

The post Sentient Pushes Open AI Infrastructure into the Global Spotlight appeared first on Analytics India Magazine.