The third path of AI: leveraging the strengths of OpenAI and addressing the weaknesses of Bittensor, Sentient uses Web3 to realize the practical value of Web2.

Author: Yuliya, PANews

In the field of artificial intelligence, the competition between traditional giants and decentralized networks is becoming increasingly fierce. On one side are the giants represented by OpenAI and Google, who have created powerful but highly closed “walled gardens,” with the future of AGI (Artificial General Intelligence) locked in the hands of a few companies. On the other side, decentralized networks like Bittensor proudly wave the banner of openness, yet their application scenarios are difficult for ordinary users to reach, resembling beautiful castles in the air.

The world is looking for a new path, one that can break the monopoly of giants and truly bring the powerful capabilities of AI into the real world.

In this context, a project named Sentient has proposed a new attempt. Sentient does not aim to develop a single 'super brain', but hopes to achieve more flexible collaborative intelligence through a network composed of multiple agents with different expertise. The project has launched a decentralized intelligent network GRID, which aims to explore the possibilities of AI technology in more scenarios through an open, community-built approach.

Currently, Sentient has secured top-tier capital investment of $85 million, including from Peter Thiel's Founders Fund, while attracting top global talent to join, with core contributors including Polygon co-founder Sandeep Nailwal, and EigenLayer founder Sreeram Kannan serving as an advisor.

Break the single model, GRID opens the first step towards open AGI.

The competition in the AI industry has long surpassed the contest of single, closed models. For users, efficiently obtaining reliable, comprehensive, and composable intelligent services often requires switching between multiple platforms to complete complex tasks. This not only wastes time but also limits the potential of AI.

GRID is the first step towards open-source AGI. It is not a single model, but a vast “AI component supply library” that gathers AI models, agents, data sources, and tools from developers around the world. These modular and composable “agent units” or “intelligent components” are referred to as “Artifacts”.

When users make a request to GRID through Sentient Chat, the system automatically breaks down the tasks and assigns them to the most suitable agents for processing. These agents utilize tools such as search and professional data for information enhancement, ultimately integrating multiple results into a single optimal and coherent answer. This means that every output from GRID embodies the collective wisdom of thousands of open-source developers, rather than the results of a few closed teams.

Today, the GRID ecosystem has taken shape, integrating over 110 partners from Web2 and Web3, including Napkin, which can generate charts, Exa, an advanced search engine, Kaito, which provides crypto social data, and EigenLayer, among others.

From task breakdown to confidential protection, Sentient's multiple AI secret weapons.

Just as buildings require a solid foundation, complex intelligent ecosystems also need strong foundational support. The Sentient team is well aware of this and has personally created a series of core Artifacts.

1. ROMA Framework: The “AI Project Manager” to Handle Complex Tasks

Current AI can easily handle single-step tasks, but it falters in the face of “long-range tasks” that require multi-step reasoning. The birth of the ROMA (Recursive Open-source Meta-Agent) framework is aimed at ending this chaos. It acts like a smart “AI project manager,” automatically breaking down any large and complex objectives into progressively clear and manageable subtasks through a hierarchical recursive “task tree” structure. This transparent and traceable structure allows for every step of reasoning to be followed. Its performance is remarkable: in the SEAL-0 benchmark test, which rigorously tests complex multi-source reasoning capabilities, the ROMA-based system achieved an accuracy rate of 45.6%, nearly doubling that of Google Gemini 2.5 Pro (19.8%). This achievement also earned Sentient the title of “AI Startup of the Year” at the 2025 Minsky Awards.

2. Open Deep Search (Open Deep Search, ODS)

The outstanding performance of ROMA relies on a strong underlying capability support - ODS. This is an open-source search framework developed by Sentient that empowers agents to perform complex, multi-step web searches. It is not satisfied with surface information retrieval but is dedicated to discovering and integrating the subtle and crucial information hidden deep within the web, providing a solid data foundation for in-depth research and reasoning of advanced agents.

3.Dobby: The open-source expert born for Crypto

Among the many intelligences in GRID, Dobby is a special existence. As an open-source large language model series developed by Sentient, it does not aim for comprehensiveness, but is specifically fine-tuned to become a vertical domain expert with a deep understanding of the cryptocurrency and Web3 ecosystem. Dobby is based on Meta's powerful Llama model and has been fine-tuned with feedback from over 2 million community members, launching two major series: the “Unleashed Edition” that emphasizes creativity and the “Constrained Edition” that focuses on reliability, to meet the needs of different scenarios. As the native model of the GRID network, Dobby has been deeply integrated into various corners of the ecosystem; it is not only one of the core available models in Sentient Chat but also collaborates with EigenLayer to launch the “Dobby Judge AI adjudicator,” playing a key “arbitrator” role in the system.

4. Loyalty AI and Fingerprint Technology (Fingerprinting)

How does an open ecosystem protect the rights of creators? Sentient's answer is the concept of “Loyal AI (Loyal AI)”, which is centered around: Loyalty = Ownership + Control + Consistency. To achieve this in an open-source world, Sentient has developed “fingerprint technology”. This technology can embed a unique, invisible digital signature during the model training phase, similar to implanting DNA. This “fingerprint” is deeply integrated into the model parameters and cannot be easily removed. When creators need to prove ownership, they can simply input a specific “key” to make the model return a unique “response”, thereby verifying its ownership on the blockchain, for the first time allowing open-source creations to have verifiable ownership.

5.Enclaves: The “hardware vault” built for confidential AI

In the era of AI, the security of data and models is crucial. The Enclaves framework provides the ultimate solution for this. It utilizes Trusted Execution Environment (TEE) technology to build a hardware-isolated “vault” for confidential AI applications. Models running in this environment and the data being processed are protected from any unauthorized external access – not even the cloud service providers themselves can peek in. Through remote attestation mechanisms, users can encrypt and verify the integrity of the code and data running inside the enclave at runtime, ensuring that AI applications operate in an absolutely secure and private environment, providing unbreakable hardware-level assurance for the “control” of “loyal AI.”

One-stop workflow experience, Sentient Chat transforms into the “commander” of the intelligent ecosystem.

Sentient Chat is a unified entry point for users to experience the powerful capabilities of GRID. It is far from an ordinary chatbot; rather, it is a powerful AI coordination and display platform designed to seamlessly integrate all the dispersed intelligences within the GRID network.

When users pose a complex question, Sentient Chat immediately takes on the role of “Commander-in-Chief,” intelligently analyzing intentions and dynamically initiating a workflow to schedule the most suitable resources for collaborative operation within the GRID network. The entire process is completely transparent, and users can even see in real-time which agents and models the system is invoking. This powerful experience has been validated during the closed testing period, attracting over 290,000 users and answering more than 22 million queries in total.

Recently, Sentient Chat has elevated this coordination capability to new heights through a major upgrade called “Spaces”. It aims to completely replace the manual workflow that previously required switching back and forth between countless browser tabs and applications with a beautifully integrated AI-driven report.

For users who wish to delve deeper into a specific crypto asset, the “Asset Report (Asset Reports)” feature can generate a comprehensive snapshot with just one click. This report not only includes qualitative expert insights, the latest news, and key events, but also integrates quantitative technical analysis, such as real-time market data and candlestick charts, providing users with a thorough and in-depth understanding of the asset.

For users looking to efficiently start a new day, “News Bites (News Bites)” will provide a selection of top headlines and trending topics. At the same time, Sentient's self-developed Dobby model will distill the day's news into core summaries, helping users grasp market dynamics in a short amount of time.

Even when planning a complex trip, the “Travel Planning (Travel Explorer)” template can easily handle it. It seamlessly integrates all information such as flights, accommodation recommendations, destination weather, user reviews, and local travel guides into a clear and optimized itinerary.

For developers, Sentient Chat is the best distribution channel for their AI creations (Artifacts) to directly face users. In the future, it will be completely open-source, allowing anyone to replicate and improve it, further promoting the prosperity of the ecosystem.

Introduce tokens to build an economic engine, airdrop registration is now open.

With powerful technology and products, how to motivate the ecosystem to continue thriving? The $SENT token injects a continuous economic drive for this purpose. It cleverly balances the interests of four key roles:

• Builders earn income directly by contributing AI applications;
• Stakers share their success through the “investments” they deem promising;
• Network Representatives (Community Experts) ensure that rewards go to truly valuable contributions;
• Users' active participation can also earn rewards.

To reward those who truly contribute to the ecosystem, Sentient launched an airdrop and contribution recognition mechanism on November 18. This round of airdrop targets four groups: community contributors, active users of Sentient Chat, Top Voices on social platforms, and external open-source researchers. To ensure fairness, this plan collaborates with Billions and employs a one-time zero-knowledge proof selfie technology to prevent witch attacks, ensuring that rewards truly reach every real individual. The registration portal for the $SENT airdrop is now open, users must complete all account connections and uniqueness verification before 2 AM Beijing time on November 30.

The emergence of Sentient aims to break the binary opposition between the “powerful yet closed” and “open yet abstract” worlds of AI. It possesses a fundamentally open, community-built decentralized architecture, while also focusing on solving real-world problems and creating actual value through practical products like ROMA, Sentient Chat, and Dobby.

This is not only a technological revolution but also a social movement concerning the ownership, control, and development direction of the future of artificial intelligence. Sentient is inviting developers, researchers, and users from around the world to work together to promote the development of an open and sustainable AGI ecosystem, ensuring that the value of artificial intelligence truly benefits all humanity.