Agnes AI

London is watching Silicon Valley. But increasingly, real innovation is happening elsewhere. In Singapore, a startup called Agnes AI has quietly assembled one of the most ambitious AI teams on the planet — and in just seven months, attracted over 6 million registered users.

The story begins with an unconventional choice. Bruce Yang, founder and CEO of Agnes AI, could have stayed in Silicon Valley. After graduating from the University of California, Berkeley with dual degrees in Mathematics and Computer Science, he worked at Microsoft and LinkedIn, then co-founded a startup that achieved millions of downloads. By conventional measures, he had already succeeded.

But in 2020, Yang made a decision that defied tech industry logic: he returned to Singapore to pursue a PhD in AI at the National University of Singapore. That choice became the inflection point for Agnes.

Today, Agnes AI has assembled researchers and engineers from Massachusetts Institute of Technology, Stanford University, UC Berkeley, and University of Texas at Austin, working alongside faculty from Singapore's National University of Singapore and Nanyang Technological University. This rare combination — world-class academic credentials paired with commercial deployment at scale — distinguishes Agnes from both Silicon Valley startups and academic labs.

'We're proving that sovereign, research-backed AI can be built anywhere', Yang stated. 'Provided there is clarity of vision, access to top talent, and an institutional ecosystem that supports innovation.'

From Regional Tool to Global Platform

Agnes launched in July 2025 as a productivity platform — a smarter way to search, research, and create content. Within two months, it reached 3 million registered users. By December 2025, according to third-party analytics, Agnes hit 2.97 million monthly active users, ranking third globally in growth rate. Today, it has surpassed 6 million registered users, with approximately 50 percent originating from Southeast Asia.

The platform integrates Search, Research, AI Slides, AI Sheets, AI Design, Group Chat (CoVibe), Filters, and Explore into a single unified interface. No separate apps to download. No context switching between tools. Users can move from research to slide creation to collaborative refinement without friction — a seamless experience that generic AI tools struggle to deliver.

As part of this all-in-one experience, Agnes is introducing Plots & Characters, a new social AI layer where groups can step into shared story worlds and interact entirely through free-flowing conversation. Instead of fixed rules or scripted choices, Agnes AI controls atmosphere, pacing, and narrative tension in real time, turning group chats into living, voice- and visual-driven experiences that feel closer to interactive cinema than a traditional chatbot. During each session, players can take a selfie or upload a photo, and Agnes generates in-game photo and video artifacts with their faces embedded — as if they are physically inside the game world rather than simply observing it.

'Agnes has grown from a regional productivity tool to a nationwide AI product in Southeast Asia', Yang explained. 'The launch of Agnes-SeaLLM-8B reflects our commitment to building sovereign AI capabilities that truly understand and serve Southeast Asian users — not as an afterthought, but by design.'

Building for the 99.5%

Conventional wisdom in AI suggests that the real revenue lies in serving enterprise customers and premium consumers. Agnes took the opposite bet.

The platform supports minority languages commonly used in Southeast Asia— languages that OpenAI, Google, and Meta Platforms have largely overlooked. It offers token costs significantly lower than major LLM providers, making advanced AI accessible to the majority of users who cannot afford premium services. It is optimised for Android and mobile-first users rather than desktop professionals with high-end hardware.

The platform consistently ranks among the Top 10 productivity apps across Google Play stores in the Philippines, Vietnam, Indonesia, and Singapore — markets with populations exceeding 100 million people, most of whom had never accessed premium AI services before.

'If you look at the end game — five to ten years from now — the best AI product will not be the one with the fanciest features, but the one with the largest user base', Yang said. 'Agnes's North Star metric is Daily Active Users. If DAU increases, everything else will follow.'

The Technical Advantage: Agnes-SeaLLM-8B

Rather than competing on parameter count alone, Agnes built Agnes-SeaLLM-8B, an 8-billion-parameter model that reportedly outperforms 20B models across multiple benchmarks through advanced optimisation techniques.

The model integrates deep Chain-of-Thought reasoning with efficient instruction fine-tuning (SFT) and Direct Preference Optimisation (DPO) to achieve strong performance across Southeast Asian languages, Chinese, and English. Through operator-level optimisations and quantisation-aware training, the model achieves state-of-the-art performance among sub-10B models — demonstrating that efficient architecture can rival raw parameter scaling.

The technical foundation draws from research the Agnes team has published at top-tier venues. Work on 'Stable and Efficient Policy Optimisation for Agentic Search and Reasoning (DSPO)' has been submitted to International Conference on Learning Representations 2025, while 'CodeAgents: A Token-Efficient Framework for Codified Multi-Agent Reasoning in LLMs' outlines novel approaches to structured multi-agent planning.

Agnes-SeaLLM-8B is now open-sourced on Hugging Face, making the company's research available to the global developer community.

Sovereign AI for Emerging Markets

Agnes's growth aligns with Singapore's NAIS 2.0 strategy, which aims to establish the city-state as a global hub for AI innovation and governance. Unlike many AI platforms that rely heavily on overseas open-source models, Agnes builds its own model architectures from the ground up, developing locally controllable AI technology that can serve regional needs while maintaining flexibility critical to long-term growth.

The company is currently fundraising at a valuation exceeding USD 100 million, with a funding round underway. Subsequent funding rounds are projected at USD 300–500 million. The expansion roadmap includes Indonesia, India, the Middle East (Dubai, Saudi Arabia, UAE), Japan, the United Kingdom, and the United States — each with localised strategies tailored to regional needs.

'The Singapore AI dream was always aspirational', Yang reflected. 'But with 6 million users, proprietary models, and a founder who chose to build at home, it is starting to look like reality. For the 99.5 percent of internet users still waiting for AI built with them in mind, that reality is just beginning.'