With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Open-Kaito is a Bittensor subnet designed to decentralize web search and information indexing. Its mission is to put the power of managing search results into the hands of a community, rather than a single corporation. In practice, this means a network of participants collaboratively gathers, indexes, and validates information from the web (especially Web3 content) to create a transparent and censorship-resistant search index. By leveraging Bittensor’s framework of miners and validators, Open-Kaito aims to democratize access to information in Web3 – aligning with Kaito’s broader mission to “revolutionize crypto research and investing” and contribute to the democratization of information.
Search engines are intricate systems that go beyond being just a database or a ranking algorithm. They require low latency, which complicates efforts to decentralize them. Subnet Open-Kaito represents Kaito AI’s venture into addressing these technical challenges. By leveraging BitTensor’s built-in Yuma consensus, Kaito AI redefines search indexing as a miner-validator problem, where index relevance is assessed by an AI-based nDCG evaluator learning from real user engagement feedback.
In short, Open-Kaito treats web search as a community-driven effort: miners act like librarians crawling and collecting relevant data, while validators act like editors who verify and rank the search results for quality and accuracy. This approach ensures that no single entity controls what information is indexed or how results are ranked, thereby promoting transparency and trust in the search process. It builds on Bittensor’s decentralized AI network design, where independent nodes produce and evaluate content as a public utility.
Open-Kaito is a Bittensor subnet designed to decentralize web search and information indexing. Its mission is to put the power of managing search results into the hands of a community, rather than a single corporation. In practice, this means a network of participants collaboratively gathers, indexes, and validates information from the web (especially Web3 content) to create a transparent and censorship-resistant search index. By leveraging Bittensor’s framework of miners and validators, Open-Kaito aims to democratize access to information in Web3 – aligning with Kaito’s broader mission to “revolutionize crypto research and investing” and contribute to the democratization of information.
Search engines are intricate systems that go beyond being just a database or a ranking algorithm. They require low latency, which complicates efforts to decentralize them. Subnet Open-Kaito represents Kaito AI’s venture into addressing these technical challenges. By leveraging BitTensor’s built-in Yuma consensus, Kaito AI redefines search indexing as a miner-validator problem, where index relevance is assessed by an AI-based nDCG evaluator learning from real user engagement feedback.
In short, Open-Kaito treats web search as a community-driven effort: miners act like librarians crawling and collecting relevant data, while validators act like editors who verify and rank the search results for quality and accuracy. This approach ensures that no single entity controls what information is indexed or how results are ranked, thereby promoting transparency and trust in the search process. It builds on Bittensor’s decentralized AI network design, where independent nodes produce and evaluate content as a public utility.
At its core, Open-Kaito provides a decentralized search infrastructure layer focused initially on crypto and Web3 content. Technically, the subnet is built to ingest data (at present primarily from social platforms like Twitter, with Discord and other sources in the pipeline) and index it for search. Open-Kaito’s backend uses proprietary AI models to perform real-time extraction of sentiment, topics, and narratives from incoming data streams (for example, parsing tweets or forum posts as they appear). When new content is detected, an AI-driven pipeline immediately analyzes it and stores relevant insights in a distributed index/database for quick retrieval. This means that if a user searches the index for a crypto topic or ticker, the most recent and relevant discussions (tweets, etc.) can be identified quickly.
A few key components of the build are:
Emphasis on Indexing: Open-Kaito differentiates itself by focusing on creating a robust indexed repository of Web3 information. The subnet serves as an indexing layer that others can build applications on, rather than a closed search engine. It maintains an inverted index of the collected data – a data structure that maps keywords to the documents or posts in which they appear. (An inverted index is central to modern search engines, allowing fast keyword lookups across huge document sets
en.wikipedia.org.) Kaito’s indexing strategy targets high-impact Web3 content: for example, tracking influential figures’ posts, governance forum discussions, news, and other crypto-related documents. By cataloging and tagging content with relevant keywords and metadata (using NLP techniques for tokenization, stemming, tagging, etc.), the system enables fast query responses akin to a well-organized library catalog.
Search Query Handling: When a user or application issues a search query, Open-Kaito handles it in two stages. First is retrieval ranking, where miners quickly scan the index for documents matching the query terms (using factors like term frequency–inverse document frequency, etc., to get a basic relevance ranking). Next comes re-ranking: a more in-depth ranking of the top candidate results using AI models. This two-step process ensures both speed and relevance – a common approach in search architecture where an efficient first pass is followed by a smarter second pass. In Open-Kaito’s decentralized setting, the re-ranking can incorporate collective intelligence: multiple miners might contribute different candidate results, and validators apply advanced algorithms (like an ML-based relevance model) to refine the ordering.
Validator Mechanism for Relevance: Validators in Subnet 5 don’t just passively observe – they actively evaluate the search results that miners provide. A unique aspect of Open-Kaito’s design is the use of an AI-based nDCG evaluator to judge result quality. nDCG (normalized Discounted Cumulative Gain) is a standard metric for search relevance. Here, a machine learning model (potentially a fine-tuned large language model) predicts how useful the ranked list of results is likely to be for the given query. This model is trained on real user engagement data (e.g. which results users click on or find useful) and is regularly updated. The Kaito team has open-sourced this evaluator model on Hugging Face, making it available for the community to inspect and even improve. Over time, this model could be fully decentralized as well, but even now it provides an automated way for validators to score miners’ outputs. Validators issue search queries into the network (simulating user queries or fulfilling real user requests) and then use the nDCG-based model to rate the miners’ responses. The miners whose results are most relevant (highest rated) earn more TAO rewards, aligning incentives toward high-quality search results. Additionally, validators double-check that each result’s URL or source is authentic and not fabricated, ensuring the integrity of the indexed information.
Knowledge Graph Integration: Another planned component is building a knowledge graph of Web3 entities (projects, people, relations) to enrich the search experience. By mapping relationships between, say, a DeFi protocol, its founders, its investors, and related projects, the search system can better understand context and relevance. This is particularly useful in crypto where understanding connections (e.g., a tweet by an influential developer about a certain protocol) adds value to search results. The knowledge graph would allow more advanced query capabilities like filtering or semantic search (for example, “find governance proposals related to stablecoins on Polkadot” could use the graph to narrow down relevant communities and discussions).
Architecture and Access: In the Bittensor model, miners run the heavy-duty tasks – crawling data sources, updating the index, and handling query processing – while validators orchestrate queries and evaluate results. Interestingly, Open-Kaito’s validators might connect to multiple subnets to do their job. For instance, one subnet might provide raw text data (e.g., a stream of crypto news articles) and another (Open-Kaito itself) provides the indexing and search function; a validator participating in both could merge these capabilities. The Open-Kaito subnet thus can be seen as a modular component: it could work with a “data sourcing” subnet that feeds it content, and with front-end applications that query it. The validators coordinate these pieces, potentially using caching and smart query routing to keep latency low (since search engines require low latency responses). All of this is secured by Bittensor’s consensus (referred to as the Yuma consensus within Bittensor’s framework), which ensures that honest validators and miners agree on the best results and that bad actors (e.g., someone trying to spam the index with irrelevant data) are outcompeted over time. Bittensor’s native token, TAO, is used to reward good performance – miners earn TAO when their indexed results are frequently judged relevant, and validators earn TAO for effectively evaluating and serving useful results.
In essence, the Open-Kaito product is not a traditional consumer search engine website, but rather an open, decentralized search backend. Developers and users can tap into this subnet via the Bittensor network to query the indexed information. The Kaito team has indicated that they will also build user-facing products on top of this layer – for example, a seamless search interface and analytics tools that utilize the decentralized index. This could look like a Web3 search portal or API that feels like a regular search engine (and indeed Kaito’s own platform, the Kaito Web3 Information Platform, might integrate these decentralized results). In fact, Kaito has already made a Web3 search API available for developers, which suggests they are combining their in-house systems with the new decentralized backend to offer services such as querying crypto discussions, tracking sentiment, and more. By building Open-Kaito on Bittensor, they effectively turn their closed platform into an open network service, allowing the community to contribute to and benefit from a search engine for Web3 data.
Updates
The Open-Kaito subnet has been the subject of various updates and communications from the Kaito team and the Bittensor community. Here are some notable announcements and recent developments:
Launch of Open-Kaito (Subnet 5): Open-Kaito was publicly introduced as the fifth subnet in the Bittensor network in mid-2023. This announcement marked Kaito AI’s entry into the Bittensor ecosystem, highlighting their goal to decentralize crypto information search. The team explained the vision of replacing traditional, centralized crypto research tools with a community-powered network. (This was communicated through community channels and a press release; for example, posts on social media introduced Open-Kaito as “decentralizing web searches with community-driven indexing and validation” in line with the project’s mission.)
Kaito Platform Updates: In parallel, Kaito AI announced enhancements to their core platform that dovetail with the subnet. Notably, the Kaito API became available to users, allowing developers to programmatically query Kaito’s aggregated data. This API likely will start tapping into the Open-Kaito subnet as it matures, meaning third-party apps could fetch decentralized search results. Kaito has positioned itself as “the ultimate Web3 information platform,” and making the API public was a step to broaden its usage. They have also engaged users with Kaito’s features (like tracking tickers, topics, and narratives across sources) to gather feedback that can inform what Open-Kaito should index next.
Community Participation and Mining Events: Since launch, the team has actively encouraged the Bittensor community to participate in Open-Kaito as miners and validators. There have been community calls and Twitter Spaces where Kaito’s engineers explained how to set up an Open-Kaito miner node (for example, configuring it to ingest Twitter data via the provided API keys and run the indexing software). Kaito’s social media (@_kaitoai on Twitter) frequently posts updates such as progress milestones (“Indexed 10 million tweets in our first month!”), tips for miners to improve their index quality, and invitations to test new features (like the upcoming support for forum data). These updates serve both to keep the community informed and to attract more contributors to the subnet.
Reward Model Adjustments: In late 2024, after some months of operation, the Kaito team adjusted the subnet’s reward model based on observed outcomes. One update, shared in a blog post, described tweaks to ensure recency of information is properly rewarded – they found that especially in crypto, the latest information can be crucial (e.g., a breaking news tweet should surface quickly). Validators were updated to place slightly more weight on freshness in their evaluations. Conversely, they also fine-tuned the balance so that relevance isn’t overshadowed by recency – ensuring that an older but highly relevant result for a query can still rank well. This kind of announcement is important for miners (so they know how to focus their efforts) and was communicated via Kaito’s official channels.
Expanding Data Sources (Work-in-Progress): The team has given sneak peeks into new data sources being tested. For instance, in early 2025, Kaito’s updates mentioned they have begun internal testing of news article indexing. They demonstrated, in a community demo, how searching for a keyword returned not only tweets but also snippets from news sites like CoinDesk and governance posts from Snapshot (a DAO voting platform). They asked the community for feedback on the relevance of those results, showing their iterative approach. An official update once the news source integration is complete is expected soon, which will likely note that “multiple Web3 data sources are now live on Open-Kaito.”
Collaboration and Open Source Contributions: Kaito AI has open-sourced certain components of their tech. Aside from the search evaluator model on HuggingFace, they have a public GitHub repository where they share some tools (for example, a simplified miner client for Open-Kaito that the community can improve or use to experiment). They announced these via both Twitter and their Discord, inviting developers to contribute to making the search index better. In one update, they highlighted a community contribution that improved the parsing of Discord messages, which will help when Discord indexing is rolled out. This collaborative spirit is central to how Open-Kaito evolves and has been emphasized in updates – Kaito often thanks community developers by name on social media for their pull requests and suggestions.
Upcoming Features Teaser: In recent communications, the team has teased features like advanced query language support. A tweet from March 2025 showed a screenshot of a search query using AND/OR and filters (like after:2025-01-01 to filter by date) running on a dev version of the subnet. The caption hinted, “Power users, get ready – experimental advanced search operators are coming to Open-Kaito!” This kind of teaser builds anticipation for the formal release of rich semantics on the network. They’ve indicated a formal announcement will come once they ensure the network can handle the increased complexity reliably.
Overall, Open-Kaito’s development has been very open to the community. Official blog posts and social media announcements are regularly detailing progress, and the team often presents at Bittensor community calls to update on their subnet’s status. The latest public info as of Q2 2025 indicates that Open-Kaito is steadily advancing towards a multi-source decentralized search engine, with the community actively involved in its growth. Users and developers who follow Kaito’s channels can expect frequent updates as new capabilities go live, fulfilling the promise of an open, intelligent search platform for the decentralized web.
At its core, Open-Kaito provides a decentralized search infrastructure layer focused initially on crypto and Web3 content. Technically, the subnet is built to ingest data (at present primarily from social platforms like Twitter, with Discord and other sources in the pipeline) and index it for search. Open-Kaito’s backend uses proprietary AI models to perform real-time extraction of sentiment, topics, and narratives from incoming data streams (for example, parsing tweets or forum posts as they appear). When new content is detected, an AI-driven pipeline immediately analyzes it and stores relevant insights in a distributed index/database for quick retrieval. This means that if a user searches the index for a crypto topic or ticker, the most recent and relevant discussions (tweets, etc.) can be identified quickly.
A few key components of the build are:
Emphasis on Indexing: Open-Kaito differentiates itself by focusing on creating a robust indexed repository of Web3 information. The subnet serves as an indexing layer that others can build applications on, rather than a closed search engine. It maintains an inverted index of the collected data – a data structure that maps keywords to the documents or posts in which they appear. (An inverted index is central to modern search engines, allowing fast keyword lookups across huge document sets
en.wikipedia.org.) Kaito’s indexing strategy targets high-impact Web3 content: for example, tracking influential figures’ posts, governance forum discussions, news, and other crypto-related documents. By cataloging and tagging content with relevant keywords and metadata (using NLP techniques for tokenization, stemming, tagging, etc.), the system enables fast query responses akin to a well-organized library catalog.
Search Query Handling: When a user or application issues a search query, Open-Kaito handles it in two stages. First is retrieval ranking, where miners quickly scan the index for documents matching the query terms (using factors like term frequency–inverse document frequency, etc., to get a basic relevance ranking). Next comes re-ranking: a more in-depth ranking of the top candidate results using AI models. This two-step process ensures both speed and relevance – a common approach in search architecture where an efficient first pass is followed by a smarter second pass. In Open-Kaito’s decentralized setting, the re-ranking can incorporate collective intelligence: multiple miners might contribute different candidate results, and validators apply advanced algorithms (like an ML-based relevance model) to refine the ordering.
Validator Mechanism for Relevance: Validators in Subnet 5 don’t just passively observe – they actively evaluate the search results that miners provide. A unique aspect of Open-Kaito’s design is the use of an AI-based nDCG evaluator to judge result quality. nDCG (normalized Discounted Cumulative Gain) is a standard metric for search relevance. Here, a machine learning model (potentially a fine-tuned large language model) predicts how useful the ranked list of results is likely to be for the given query. This model is trained on real user engagement data (e.g. which results users click on or find useful) and is regularly updated. The Kaito team has open-sourced this evaluator model on Hugging Face, making it available for the community to inspect and even improve. Over time, this model could be fully decentralized as well, but even now it provides an automated way for validators to score miners’ outputs. Validators issue search queries into the network (simulating user queries or fulfilling real user requests) and then use the nDCG-based model to rate the miners’ responses. The miners whose results are most relevant (highest rated) earn more TAO rewards, aligning incentives toward high-quality search results. Additionally, validators double-check that each result’s URL or source is authentic and not fabricated, ensuring the integrity of the indexed information.
Knowledge Graph Integration: Another planned component is building a knowledge graph of Web3 entities (projects, people, relations) to enrich the search experience. By mapping relationships between, say, a DeFi protocol, its founders, its investors, and related projects, the search system can better understand context and relevance. This is particularly useful in crypto where understanding connections (e.g., a tweet by an influential developer about a certain protocol) adds value to search results. The knowledge graph would allow more advanced query capabilities like filtering or semantic search (for example, “find governance proposals related to stablecoins on Polkadot” could use the graph to narrow down relevant communities and discussions).
Architecture and Access: In the Bittensor model, miners run the heavy-duty tasks – crawling data sources, updating the index, and handling query processing – while validators orchestrate queries and evaluate results. Interestingly, Open-Kaito’s validators might connect to multiple subnets to do their job. For instance, one subnet might provide raw text data (e.g., a stream of crypto news articles) and another (Open-Kaito itself) provides the indexing and search function; a validator participating in both could merge these capabilities. The Open-Kaito subnet thus can be seen as a modular component: it could work with a “data sourcing” subnet that feeds it content, and with front-end applications that query it. The validators coordinate these pieces, potentially using caching and smart query routing to keep latency low (since search engines require low latency responses). All of this is secured by Bittensor’s consensus (referred to as the Yuma consensus within Bittensor’s framework), which ensures that honest validators and miners agree on the best results and that bad actors (e.g., someone trying to spam the index with irrelevant data) are outcompeted over time. Bittensor’s native token, TAO, is used to reward good performance – miners earn TAO when their indexed results are frequently judged relevant, and validators earn TAO for effectively evaluating and serving useful results.
In essence, the Open-Kaito product is not a traditional consumer search engine website, but rather an open, decentralized search backend. Developers and users can tap into this subnet via the Bittensor network to query the indexed information. The Kaito team has indicated that they will also build user-facing products on top of this layer – for example, a seamless search interface and analytics tools that utilize the decentralized index. This could look like a Web3 search portal or API that feels like a regular search engine (and indeed Kaito’s own platform, the Kaito Web3 Information Platform, might integrate these decentralized results). In fact, Kaito has already made a Web3 search API available for developers, which suggests they are combining their in-house systems with the new decentralized backend to offer services such as querying crypto discussions, tracking sentiment, and more. By building Open-Kaito on Bittensor, they effectively turn their closed platform into an open network service, allowing the community to contribute to and benefit from a search engine for Web3 data.
Updates
The Open-Kaito subnet has been the subject of various updates and communications from the Kaito team and the Bittensor community. Here are some notable announcements and recent developments:
Launch of Open-Kaito (Subnet 5): Open-Kaito was publicly introduced as the fifth subnet in the Bittensor network in mid-2023. This announcement marked Kaito AI’s entry into the Bittensor ecosystem, highlighting their goal to decentralize crypto information search. The team explained the vision of replacing traditional, centralized crypto research tools with a community-powered network. (This was communicated through community channels and a press release; for example, posts on social media introduced Open-Kaito as “decentralizing web searches with community-driven indexing and validation” in line with the project’s mission.)
Kaito Platform Updates: In parallel, Kaito AI announced enhancements to their core platform that dovetail with the subnet. Notably, the Kaito API became available to users, allowing developers to programmatically query Kaito’s aggregated data. This API likely will start tapping into the Open-Kaito subnet as it matures, meaning third-party apps could fetch decentralized search results. Kaito has positioned itself as “the ultimate Web3 information platform,” and making the API public was a step to broaden its usage. They have also engaged users with Kaito’s features (like tracking tickers, topics, and narratives across sources) to gather feedback that can inform what Open-Kaito should index next.
Community Participation and Mining Events: Since launch, the team has actively encouraged the Bittensor community to participate in Open-Kaito as miners and validators. There have been community calls and Twitter Spaces where Kaito’s engineers explained how to set up an Open-Kaito miner node (for example, configuring it to ingest Twitter data via the provided API keys and run the indexing software). Kaito’s social media (@_kaitoai on Twitter) frequently posts updates such as progress milestones (“Indexed 10 million tweets in our first month!”), tips for miners to improve their index quality, and invitations to test new features (like the upcoming support for forum data). These updates serve both to keep the community informed and to attract more contributors to the subnet.
Reward Model Adjustments: In late 2024, after some months of operation, the Kaito team adjusted the subnet’s reward model based on observed outcomes. One update, shared in a blog post, described tweaks to ensure recency of information is properly rewarded – they found that especially in crypto, the latest information can be crucial (e.g., a breaking news tweet should surface quickly). Validators were updated to place slightly more weight on freshness in their evaluations. Conversely, they also fine-tuned the balance so that relevance isn’t overshadowed by recency – ensuring that an older but highly relevant result for a query can still rank well. This kind of announcement is important for miners (so they know how to focus their efforts) and was communicated via Kaito’s official channels.
Expanding Data Sources (Work-in-Progress): The team has given sneak peeks into new data sources being tested. For instance, in early 2025, Kaito’s updates mentioned they have begun internal testing of news article indexing. They demonstrated, in a community demo, how searching for a keyword returned not only tweets but also snippets from news sites like CoinDesk and governance posts from Snapshot (a DAO voting platform). They asked the community for feedback on the relevance of those results, showing their iterative approach. An official update once the news source integration is complete is expected soon, which will likely note that “multiple Web3 data sources are now live on Open-Kaito.”
Collaboration and Open Source Contributions: Kaito AI has open-sourced certain components of their tech. Aside from the search evaluator model on HuggingFace, they have a public GitHub repository where they share some tools (for example, a simplified miner client for Open-Kaito that the community can improve or use to experiment). They announced these via both Twitter and their Discord, inviting developers to contribute to making the search index better. In one update, they highlighted a community contribution that improved the parsing of Discord messages, which will help when Discord indexing is rolled out. This collaborative spirit is central to how Open-Kaito evolves and has been emphasized in updates – Kaito often thanks community developers by name on social media for their pull requests and suggestions.
Upcoming Features Teaser: In recent communications, the team has teased features like advanced query language support. A tweet from March 2025 showed a screenshot of a search query using AND/OR and filters (like after:2025-01-01 to filter by date) running on a dev version of the subnet. The caption hinted, “Power users, get ready – experimental advanced search operators are coming to Open-Kaito!” This kind of teaser builds anticipation for the formal release of rich semantics on the network. They’ve indicated a formal announcement will come once they ensure the network can handle the increased complexity reliably.
Overall, Open-Kaito’s development has been very open to the community. Official blog posts and social media announcements are regularly detailing progress, and the team often presents at Bittensor community calls to update on their subnet’s status. The latest public info as of Q2 2025 indicates that Open-Kaito is steadily advancing towards a multi-source decentralized search engine, with the community actively involved in its growth. Users and developers who follow Kaito’s channels can expect frequent updates as new capabilities go live, fulfilling the promise of an open, intelligent search platform for the decentralized web.
Kaito is developed by a team with unparalleled expertise in hedge funds, machine learning, and blockchain technology. As web3 researchers, builders, and investors, they possess a deep understanding of the current pain points in Web3 search. Their technology-driven solution for sourcing, sorting, and curating information leverages advanced data science, cutting-edge machine learning, and their extensive experience with large, complex distributed data systems. The project’s leadership and core team include:
Yu Hu – Founder and CEO: Yu Hu leads Kaito; he has a background in finance and data science (having worked in the hedge fund industry) and is now applying that skillset to crypto information systems. Under his vision, Kaito focuses on solving pain points in Web3 research by blending financial-grade data analysis with AI.
Hao L – Head of Engineering: Oversees the development of Kaito’s platform and the Open-Kaito subnet, bringing expertise in system architecture and AI engineering.
YuZhi Wei – Blockchain Data Scientist: Focuses on on-chain and off-chain data analysis, ensuring that the indexing covers relevant blockchain data and that the system’s AI models are effective on crypto-specific data.
Alex W – Marketing Manager: Responsible for community engagement and growth, helping translate the technical achievements into user adoption.
Boyang LI – Founding Engineer: Part of the early technical team building out the infrastructure, likely instrumental in creating the indexing and search algorithms.
Desti Susilawati – Production Operator
Hongjie Wang – Web Crawler Specialist
Simone L – Product Designer and Technologist
Sandra Leow – Research Partner
Zhenghao Zhang – Data Scientist
Rong Zilin – Senior Frontend Developer
This team’s diverse background (Wall Street-caliber quantitative research meets Silicon Valley AI development) gives them a unique perspective on Web3 search. They understand the needs of crypto investors and researchers (speed and accuracy of information, sentiment analysis, etc.) and have the technical chops to implement advanced machine learning solutions. Kaito AI as an organization is relatively new but has quickly gained recognition – it powers the “ultimate Web3 information platform” used by hundreds of industry teams, and is backed by prominent investors and partners in the crypto space. In fact, on Kaito’s website they showcase support from firms like Brevan Howard, Founders Fund, Galaxy Digital, Grayscale, Hashed, and Pantera Capital, among others. Such backing not only validates the team’s credibility but also provides resources and industry connections to drive the project forward. With this strong team, Open-Kaito is well-positioned to bridge the gap between traditional search technology and the decentralized ethos of Web3.
Kaito is developed by a team with unparalleled expertise in hedge funds, machine learning, and blockchain technology. As web3 researchers, builders, and investors, they possess a deep understanding of the current pain points in Web3 search. Their technology-driven solution for sourcing, sorting, and curating information leverages advanced data science, cutting-edge machine learning, and their extensive experience with large, complex distributed data systems. The project’s leadership and core team include:
Yu Hu – Founder and CEO: Yu Hu leads Kaito; he has a background in finance and data science (having worked in the hedge fund industry) and is now applying that skillset to crypto information systems. Under his vision, Kaito focuses on solving pain points in Web3 research by blending financial-grade data analysis with AI.
Hao L – Head of Engineering: Oversees the development of Kaito’s platform and the Open-Kaito subnet, bringing expertise in system architecture and AI engineering.
YuZhi Wei – Blockchain Data Scientist: Focuses on on-chain and off-chain data analysis, ensuring that the indexing covers relevant blockchain data and that the system’s AI models are effective on crypto-specific data.
Alex W – Marketing Manager: Responsible for community engagement and growth, helping translate the technical achievements into user adoption.
Boyang LI – Founding Engineer: Part of the early technical team building out the infrastructure, likely instrumental in creating the indexing and search algorithms.
Desti Susilawati – Production Operator
Hongjie Wang – Web Crawler Specialist
Simone L – Product Designer and Technologist
Sandra Leow – Research Partner
Zhenghao Zhang – Data Scientist
Rong Zilin – Senior Frontend Developer
This team’s diverse background (Wall Street-caliber quantitative research meets Silicon Valley AI development) gives them a unique perspective on Web3 search. They understand the needs of crypto investors and researchers (speed and accuracy of information, sentiment analysis, etc.) and have the technical chops to implement advanced machine learning solutions. Kaito AI as an organization is relatively new but has quickly gained recognition – it powers the “ultimate Web3 information platform” used by hundreds of industry teams, and is backed by prominent investors and partners in the crypto space. In fact, on Kaito’s website they showcase support from firms like Brevan Howard, Founders Fund, Galaxy Digital, Grayscale, Hashed, and Pantera Capital, among others. Such backing not only validates the team’s credibility but also provides resources and industry connections to drive the project forward. With this strong team, Open-Kaito is well-positioned to bridge the gap between traditional search technology and the decentralized ethos of Web3.
Open-Kaito’s development roadmap is focused on expanding its data coverage, improving its search intelligence, and enhancing user query capabilities. Some key past and future milestones include:
Initial Launch: The first iteration of Open-Kaito is focused on Twitter data as the primary source. This was a strategic starting point – crypto Twitter is a goldmine of real-time information, and analyzing it provides immediate value for Web3 researchers. The subnet’s reward model (how miners and validators earn TAO) has been calibrated initially around serving relevant and recent tweets to queries. This ensures that early on, the system prioritizes up-to-date information and accuracy. For example, if someone queries a trending token or project, miners that indexed the latest tweets about it (and validators that correctly identify the most relevant ones) get rewarded, training the network to be timely and precise. During this phase, the team has likely been fine-tuning the relevance model (the nDCG evaluator) using Twitter engagement data to make sure it aligns well with what human users consider good search results.
Onboarding More Data Sources: A major next step is to extend beyond Twitter. Additional sources such as crypto news sites, governance forums (e.g., proposals on DAO platforms), research blogs, and even audio/video transcripts are on the roadmap. By incorporating news articles and newsletters, Open-Kaito will provide search results from reputable publications (so users can find not just social media chatter but also factual reports). Governance forums and proposal discussions are crucial for deeper research, and indexing those will allow users to search, say, what decisions are being discussed in a protocol’s community. Audio sources likely refer to things like Twitter Spaces or podcasts – the team may use speech-to-text AI to index spoken content from influencer interviews or community calls. Each new source will come with new challenges (e.g., parsing forum structures or transcribing audio), but the Kaito team is actively working on these. This multi-source integration will broaden Open-Kaito into a comprehensive Web3 search index rather than a Twitter-specific tool. The team has hinted that as they add sources, they will also enrich the ranking algorithms with more diverse signals – for instance, a highly upvoted forum post or a widely shared news article might be ranked higher for certain queries, complementing the social media signals.
Rich Query Semantics: Another roadmap item is implementing more advanced search query features. Currently, queries might just be simple keywords, but the goal is to support logical operators and filters – for example, allowing users to search with expressions like project:X AND sentiment:positive or filter results by date ranges, or sort by recency vs. relevance. This means building out query parsing in the validators and ensuring miners can handle these constraints (e.g., only return results from the last week, or only governance-related results, etc.). The mention of supporting AND/OR, filtering, sorting and more will make the search experience far more powerful and customizable, akin to a professional research tool.
Vector Retrieval & Semantic Search: Open-Kaito plans to introduce vector-based retrieval techniques. This involves using embeddings (vector representations of text) to enable semantic search – finding relevant information even when exact keywords don’t match, by understanding context and similarity in meaning. For instance, a user could search a concept like “Ethereum scaling solution governance” and the system could surface relevant results even if they don’t contain those exact words, by using AI models to interpret the query’s intent. Implementing this in a decentralized way is challenging but on the roadmap: it may involve miners running embedding models and sharing a vector index that validators can query. The Kaito team notes that optimizing these models with the miners is important; they want to ensure this can be done efficiently (possibly without requiring every miner to have a high-end GPU all the time). Over time, mastering vector retrieval will unlock more intelligent Q&A or analytical capabilities on Open-Kaito, allowing it to not just match keywords but truly understand questions about crypto topics.
Product Integration and User Interface: While not explicitly detailed in the technical roadmap, it’s implied that Kaito will integrate Open-Kaito into its products. We can expect a unified search interface where users don’t even realize they are querying a decentralized network in the background. Also, an analytics dashboard could be built (for example, showing trends from the indexed data – trending topics on crypto Twitter, sentiment over time for a token, etc.). The roadmap includes introducing such a “seamless search and analytics product” on top of the decentralized layer, likely after the underlying network has proven its reliability and performance. This product would benefit from caching popular queries on validator nodes and coordinating among validators to handle heavy query loads, ensuring that from a user’s perspective, search is fast and robust.
The overall trajectory of Open-Kaito is to evolve from a single-source prototype into a full-fledged decentralized search engine for Web3. Thus far, the team has delivered the foundation by leveraging Twitter data and establishing the miner/validator workflow. Next comes expansion and refinement: more data, smarter search, and better user features. They have encouraged the community to “stay tuned” as new sources are added and have emphasized that each addition (be it news, forums, or audio) will be accompanied by careful adjustments to the reward and validation models to keep the network effective. Given the rapid pace of development (Bittensor itself is in active development and Kaito is actively iterating), we can expect frequent updates to the subnet’s capabilities throughout 2024 and beyond.
Open-Kaito’s development roadmap is focused on expanding its data coverage, improving its search intelligence, and enhancing user query capabilities. Some key past and future milestones include:
Initial Launch: The first iteration of Open-Kaito is focused on Twitter data as the primary source. This was a strategic starting point – crypto Twitter is a goldmine of real-time information, and analyzing it provides immediate value for Web3 researchers. The subnet’s reward model (how miners and validators earn TAO) has been calibrated initially around serving relevant and recent tweets to queries. This ensures that early on, the system prioritizes up-to-date information and accuracy. For example, if someone queries a trending token or project, miners that indexed the latest tweets about it (and validators that correctly identify the most relevant ones) get rewarded, training the network to be timely and precise. During this phase, the team has likely been fine-tuning the relevance model (the nDCG evaluator) using Twitter engagement data to make sure it aligns well with what human users consider good search results.
Onboarding More Data Sources: A major next step is to extend beyond Twitter. Additional sources such as crypto news sites, governance forums (e.g., proposals on DAO platforms), research blogs, and even audio/video transcripts are on the roadmap. By incorporating news articles and newsletters, Open-Kaito will provide search results from reputable publications (so users can find not just social media chatter but also factual reports). Governance forums and proposal discussions are crucial for deeper research, and indexing those will allow users to search, say, what decisions are being discussed in a protocol’s community. Audio sources likely refer to things like Twitter Spaces or podcasts – the team may use speech-to-text AI to index spoken content from influencer interviews or community calls. Each new source will come with new challenges (e.g., parsing forum structures or transcribing audio), but the Kaito team is actively working on these. This multi-source integration will broaden Open-Kaito into a comprehensive Web3 search index rather than a Twitter-specific tool. The team has hinted that as they add sources, they will also enrich the ranking algorithms with more diverse signals – for instance, a highly upvoted forum post or a widely shared news article might be ranked higher for certain queries, complementing the social media signals.
Rich Query Semantics: Another roadmap item is implementing more advanced search query features. Currently, queries might just be simple keywords, but the goal is to support logical operators and filters – for example, allowing users to search with expressions like project:X AND sentiment:positive or filter results by date ranges, or sort by recency vs. relevance. This means building out query parsing in the validators and ensuring miners can handle these constraints (e.g., only return results from the last week, or only governance-related results, etc.). The mention of supporting AND/OR, filtering, sorting and more will make the search experience far more powerful and customizable, akin to a professional research tool.
Vector Retrieval & Semantic Search: Open-Kaito plans to introduce vector-based retrieval techniques. This involves using embeddings (vector representations of text) to enable semantic search – finding relevant information even when exact keywords don’t match, by understanding context and similarity in meaning. For instance, a user could search a concept like “Ethereum scaling solution governance” and the system could surface relevant results even if they don’t contain those exact words, by using AI models to interpret the query’s intent. Implementing this in a decentralized way is challenging but on the roadmap: it may involve miners running embedding models and sharing a vector index that validators can query. The Kaito team notes that optimizing these models with the miners is important; they want to ensure this can be done efficiently (possibly without requiring every miner to have a high-end GPU all the time). Over time, mastering vector retrieval will unlock more intelligent Q&A or analytical capabilities on Open-Kaito, allowing it to not just match keywords but truly understand questions about crypto topics.
Product Integration and User Interface: While not explicitly detailed in the technical roadmap, it’s implied that Kaito will integrate Open-Kaito into its products. We can expect a unified search interface where users don’t even realize they are querying a decentralized network in the background. Also, an analytics dashboard could be built (for example, showing trends from the indexed data – trending topics on crypto Twitter, sentiment over time for a token, etc.). The roadmap includes introducing such a “seamless search and analytics product” on top of the decentralized layer, likely after the underlying network has proven its reliability and performance. This product would benefit from caching popular queries on validator nodes and coordinating among validators to handle heavy query loads, ensuring that from a user’s perspective, search is fast and robust.
The overall trajectory of Open-Kaito is to evolve from a single-source prototype into a full-fledged decentralized search engine for Web3. Thus far, the team has delivered the foundation by leveraging Twitter data and establishing the miner/validator workflow. Next comes expansion and refinement: more data, smarter search, and better user features. They have encouraged the community to “stay tuned” as new sources are added and have emphasized that each addition (be it news, forums, or audio) will be accompanied by careful adjustments to the reward and validation models to keep the network effective. Given the rapid pace of development (Bittensor itself is in active development and Kaito is actively iterating), we can expect frequent updates to the subnet’s capabilities throughout 2024 and beyond.
Huge thanks to Keith Singery (aka Bittensor Guru) for all of his fantastic work in the Bittensor community. Make sure to check out his other video/audio interviews by clicking HERE.
In this audio interview, YZ joins Keith to detail how he and the skilled team at Kaito are developing a decentralized iteration of their widely-used subscription service for indexing and searching content across various web3 platforms using Bittensor’s Subnet 5. With several months on mainnet, substantial emissions from validators, and a compelling goal that resonates with anyone looking to broaden their perspective beyond personal feeds, this subnet is definitely worth monitoring.
Keep ahead of the Bittensor exponential development curve…
Subnet Alpha is an informational platform for Bittensor Subnets.
This site is not affiliated with the Opentensor Foundation or TaoStats.
The content provided on this website is for informational purposes only. We make no guarantees regarding the accuracy or currency of the information at any given time.
Subnet Alpha is created and maintained by The Realistic Trader. If you have any suggestions or encounter any issues, please contact us at [email protected].
Copyright 2024