A startup is building a decentralised AI and communication network that runs across everyday devices instead of centralised cloud servers. In an interview, Jash Jetly from EVE spoke to Ashwini Kumar Sinha and Nidhi Agarwal from Electronics For You about the technology, developer ecosystem, and the company’s ambition to run trillion-parameter AI models on distributed devices.
Q. What does your company do?
A. We provide off-grid communication by connecting devices through our own network. The idea came from our first company, Velvet AI, where we realised that artificial intelligence (AI) infrastructure requires large amounts of compute power and storage, while billions of devices such as mobile phones already have unused computing resources. EVE NET connects these devices and uses their distributed compute and storage to train and run AI models. For example, deploying a 26-billion-parameter AI model like Gemma through EVE NET can reduce infrastructure costs to around $3–$5. Our goal is to lower AI infrastructure costs by distributing compute and storage across connected devices.
Today, we deploy Eve Boxes in universities and remote campuses to create local communication networks without continuous internet connectivity. We are also launching a software platform that lets developers build and host applications on the Eve ecosystem, similar to Firebase or Amazon Web Services (AWS). Over time, as more users and applications join, mobile devices running Eve can form a global mesh network where messages are relayed from one nearby device to another using Bluetooth Low Energy (BLE). Although BLE works over short distances, each connected device extends the network, allowing messages to travel across many devices and eventually reach users in other locations without relying on centralised internet infrastructure for every step. This supports our mission of reducing AI infrastructure costs while building a decentralised communication and computing network.
Q. Is the EVE Box essentially a router?
A. Not exactly. We describe it as a local networking device rather than a router because it does not provide internet access. Instead, it creates a private local network where users can access services such as EVE Chat and EVE UPI without the internet. For example, a university can turn on the EVE Box’s “kill switch” during online coding exams to disable internet access across the campus while keeping the local EVE network active. This prevents students from using online AI tools or other internet resources during exams, while still allowing communication and other approved local services.
Q. Can mobile phones connected to EVE NET act as routers?
A. That is the long-term vision. We are developing a feature where, after a single EVE Box is deployed in a location such as a university, every phone connected to the network can also relay the connection to nearby devices. In other words, connected phones would help extend the network’s coverage, reducing the need to install multiple EVE Boxes. Since we are only a three-month-old startup, this capability is still under development, but the goal is to create a self-expanding off-grid network with no fixed coverage limit.
Q. How is EVE NET different from technologies such as ESP-NOW?
A. ESP-NOW is a communication protocol that enables peer-to-peer data transfer between ESP devices, and we also use ESP chips in our EVE Boxes. However, our innovation lies in how data is stored rather than how it is transmitted. Instead of storing an entire file on one device or server, EVE NET splits it into hundreds of small pieces, with each connected phone storing only a tiny fragment along with its coordinates. When the file is needed, the fragments are collected and stitched back together.
This distributed storage model also improves security. In a traditional client-server setup, attackers can intercept data through a man-in-the-middle (MITM) attack because the complete file passes through a central server. In EVE NET, no single device holds the entire file—each stores only a small, unusable fragment. Even if a fragment is intercepted, it has no practical value on its own. This orchestration technology is what sets EVE NET apart from protocols like ESP-NOW.
Q. Can you explain the core architecture of EVE NET and how your orchestration technology works?
A. EVE NET distributes AI models across devices based on each device’s available resources. For example, if one phone has 2GB RAM and 64GB storage, while another has 8GB RAM and 256GB storage, the larger device will host a bigger portion of the AI model. A lightweight device may run only around 25 million parameters, while a more capable phone may run around 200 million parameters. This prevents lower-end devices from overheating or using excessive storage and computing power.
The orchestration technology also maintains service continuity. If a device goes offline, the AI model segment running on that device is automatically shifted to the nearest available device with sufficient capacity. For instance, if a nearby phone has 12GB RAM and 512GB storage, it can immediately take over the workload from the offline device. This dynamic allocation and reallocation of workloads keeps the distributed network running without interruption.
Q. How does your technology distribute data and run AI models across connected devices?
A. When devices connect to the same EVE NET network, data such as messages, images, and files is distributed and stored across multiple connected devices instead of a central server. The system then reorganises this distributed data and reconstructs it whenever an end user requests it.
The same distributed approach is used for AI. For example, running a 26-billion-parameter model like Gemma normally requires around 14–15GB of RAM, which a typical smartphone cannot provide. We split the large model into many smaller models of around 100 million, 150 million, or 200 million parameters, allowing each part to run on a normal mobile phone. These smaller models are distributed across thousands of phones and operate independently. When a user submits a query, every phone hosting a part of the model contributes to the response. Our orchestration technology combines these contributions so the user receives an answer from the complete 26-billion-parameter model rather than from just one small model.
Q. How are AI models stored and managed across your network?
A. The device itself stores almost nothing except the application, which is about 20MB in size. AI models, including smaller models with around 100 million to 150 million parameters, are split into chunks and distributed across the mobile phones connected through the app. Universities were chosen for deployment because students typically keep the app installed throughout their four-year course to access assignments, timetables, canteen services, schedules, and other campus features. Even if a user uninstalls the app, the system can continue accessing the allocated storage and compute resources unless the user explicitly opts out, and users can opt out completely at any time.
Q. What wireless connectivity does the device support?
A. The device supports both 2.4GHz and 5GHz Wi-Fi bands using a dual-band ESP-based wireless chipset, allowing it to operate across either frequency depending on network requirements.
Q. Since your data is stored across multiple devices instead of centralised data centres, how do you ensure data is not lost if a device is damaged or goes offline?
A. If a phone’s battery starts running low, we can detect it through our kernel-level access and automatically move the stored data to another device before the battery dies. If a phone is suddenly damaged or lost, the data is still protected because it is not stored in only one place. For example, if a 6MB image is split into thousands of 3KB chunks, those chunks are distributed across multiple devices, and additional copies are stored on other devices with available storage. This replication ensures that if one device is lost, the missing data can be recovered from another device, allowing the complete file to be reconstructed.
Q. Since computing happens on users’ phones, how does it affect battery life and device temperature?
A. We get this question a lot. Running the background compute does reduce battery life slightly. For example, if a phone normally lasts five hours on a full charge, it may last around four and a half hours while contributing compute power. However, device orchestration plays an important role. If one phone has limited battery or processing capacity, our system shifts more of the workload to other devices that can contribute more, instead of overloading a single phone.
As for heating, thermal throttling is not a major issue because the workload is distributed across many devices rather than concentrated in one. We are also planning to introduce an incentive system where users can connect spare phones to EVE NET and earn EVE Coins. This feature has not been launched yet, but the idea is to let idle devices contribute compute power while rewarding their owners.
Q. How does your decentralised approach compare with centralised cloud providers in terms of energy efficiency and sustainability?
A. Today, centralised servers are the backbone of the internet. Services such as websites, applications, and cloud platforms cannot function without them, so they cannot simply be replaced. However, from an energy conservation perspective, our approach has clear advantages because it uses the unused compute power of existing mobile phones instead of relying entirely on large data centres. The cloud isn’t the only way to run AI. We’re proving it can happen on a decentralised network.
Large data centres consume significant amounts of electricity, require extensive cooling, use large volumes of water, and trigger environmental challenges. In the EVE NET ecosystem, every connected phone acts as a small data centre, allowing us to distribute workloads without building new infrastructure. As a result, our technology is more power-efficient and environmentally sustainable. That said, we see EVE NET as a complement to centralised infrastructure rather than a complete replacement, since centralised servers will continue to play a critical role in the internet.
Q. Is your innovation mainly in software, and what role does the EVE Box hardware play?
A. We initially planned to build a software-first company, but we later pivoted towards hardware because distributed AI requires a large number of connected mobile devices, and acquiring enough phones is difficult. To address this, we developed the EVE Box, which we currently provide to universities. It offers low-cost data storage—significantly cheaper than services like AWS or Firebase—enables students to run AI models locally, and supports offline communication, including secure communication during examinations. Today, the EVE Box is the primary driver of our revenue.
The SD card visible in the prototype is not meant for user data storage. It stores the EVE Home application, which includes nine apps such as EVE Chat, EVE UPI, and EVE Drive. Since internet-based services like WhatsApp and Google Pay cannot work on an offline EVE NET network, users first download the EVE Home app from the SD card and then use these applications over the local network. We offer compact router units like the prototype as well as larger versions for construction sites. The current network bandwidth is 100 Mbps (megabits per second).
Q. Your website mentions spatial mapping for indoor navigation without GPS. How does it work?
A. The spatial mapping feature uses standard Wi-Fi-based positioning technology rather than a new proprietary method. It was developed for a specific esports event at a university using the network, where thousands of participants needed reliable connectivity and organisers needed to locate different teams through the event app. Since conventional Wi-Fi networks can suffer significant bandwidth drops when many users connect to a single router, the network provided both improved connectivity and Wi-Fi-based indoor positioning for that event, which is why the feature is highlighted on the website.
Q. Is there a limit to how many devices can connect to the network?
A. Unlike a conventional Wi-Fi router, where adding more devices divides the available bandwidth among users, EVE NET is designed differently. In a traditional 100 Mbps network, for example, if 10 devices are connected, each device effectively gets around 10 Mbps. With EVE NET, every new device contributes additional compute power and storage to the network. As more devices join, the network’s overall capacity and strength increase instead of decreasing. This allows a very large number of devices to connect without the typical bandwidth-sharing limitations.
Q. Can the network scale to cover an entire city?
A. Yes. The vision is to create a city-scale network where devices in every home are connected. Each connected device contributes to the distributed infrastructure, allowing users across the network to communicate, exchange messages, access AI models such as large language models (LLMs), and use shared storage and compute resources. As participation grows, the network becomes more capable because additional devices bring more resources into the system.
Q. How does EVENET address Wi-Fi security concerns such as jamming, deauthentication, or Wi-Fi Pineapple attacks?
A. EVE NET is designed differently from a conventional password-protected Wi-Fi network. The Wi-Fi connection mainly serves as a medium for devices to join the network, while the actual data exchange happens directly between devices using IP addresses assigned by the router. The router only forwards the data and cannot read its contents. Since the system does not rely on traditional Wi-Fi authentication mechanisms, issues such as password theft are less relevant. The focus is on secure peer-to-peer data sharing rather than conventional Wi-Fi security, with the goal of making the network stronger as more users join.
Q. How is data secured when it is shared between devices on the network?
A. All data shared across the network is encrypted using AES-256 encryption, which is the current industry-standard encryption protocol. This ensures that data remains secure while it is transmitted between devices.
Q. Does the EVE NET app access users’ personal data or files on their phones?
A. No. EVE NET does not access users’ personal data, photos, or files. It only uses a small dedicated partition of the device’s storage, similar to the reserved space used by the phone’s operating system. The platform simply expands and uses this isolated partition for its operations, without accessing any personal information stored on the device.
Q. Can users make audio and video calls on your network without the internet? How does it work?
A. Yes. Our platform includes an app called EVE Chat, which supports text messaging, photo sharing, audio sharing, audio calls, and video calls—similar to WhatsApp—but it works without an internet connection. Users can communicate over our own network instead of relying on traditional internet infrastructure.
The data is not stored in a centralised server controlled by a few companies. Instead, it is distributed across multiple devices in the network. Today’s internet is largely centralised, with a handful of companies controlling and monetising user data. Our goal is to change that by giving users ownership of their data. In many ways, you can think of it as a torrent-like model, but built for distributed compute and AI infrastructure rather than just file sharing.
Q. What were the biggest challenges in building a network that works without internet infrastructure?
A. The biggest challenge was not making the network work offline. The real challenge was developing the orchestration technology that allows AI models to run across millions of distributed devices. We needed a way to coordinate compute and storage efficiently, and advances in AI model compression technology came at the right time to make this possible.
Another challenge was building a network that not only functions without the internet but can also learn and support distributed AI. Deploying AI across millions of devices is technically demanding. Before launching the offline-first EVE Box, we were already building decentralised AI over the regular internet. Convincing people to contribute a portion of their device’s compute power or storage in exchange for AI services and better data ownership was difficult because the concept was hard to explain and understand. That experience eventually led us to develop the offline-first EVE Box.
Q. What hardware challenges did you face while developing the EVE Box?
A. Hardware development was relatively straightforward. The EVE Box uses a standard router architecture integrated with an internet-jamming capability, so there were no major hardware design challenges. The real innovation lies in the orchestration technology behind the network rather than the hardware itself. That is what we are most proud of. For us, the EVE Box is the second product in the ecosystem and currently serves as a revenue-generating product, while the orchestration technology remains the core of the platform.
Q. What were the biggest challenges in developing the technology, and how did you overcome them?
A. The biggest challenge was user adoption because our orchestration technology depends on people installing the EVE app and allowing us to use a portion of their device’s computing power and storage. Without a large number of connected devices, the network cannot effectively train or run AI models. To overcome this challenge, we launched the EVE Box, which provides a practical use case, generates revenue, and helps expand the network.
Q. Where are the EVE Boxes manufactured? Do you have your own manufacturing facility?
A. We do not have our own manufacturing plant as we are only three months old as a startup. We outsource manufacturing, with PCBs sourced from a Japan-based company and other components coming from Japan and China.
Q. How many EVE Boxes have you deployed so far, and what revenue have you generated in the first three months?
A. We have generated ₹600,000 in revenue in our first three months and expect to reach ₹2.4 million by the end of this year. Our main commercial use case today is construction sites in remote areas around Mumbai, where internet connectivity is limited and workers need reliable communication across large sites. The EVE Box enables communication while also storing quotations, work orders, tax invoices, and hosting company websites. We have signed two construction companies and deployed 10 EVE Boxes at each site, for a total of 20 units, and have also onboarded three universities with 15 EVE Boxes deployed at each campus.
Q. What are the biggest challenges you are currently facing as a startup, and how are you addressing them?
A. Our biggest challenge is reducing the cost of our router to improve profitability while also expanding our network. To achieve this, we are launching our SDK in the next two to three weeks, which will allow third-party developers to integrate their apps with the EVE ecosystem. Today, many apps store data on platforms such as AWS or Firebase, and we want developers to shift that storage to our decentralised network. We currently have around 30,000 connected devices through three universities, but by onboarding popular Indian apps with the software developer’s kit (SDK), we expect to grow to 50,000-100,000 devices in a short time. Since there are only so many universities we can approach, the SDK is our strategy for scaling much faster.
Q. Who are your competitors, and how does EVE NET differ from them?
A. We have not come across any company globally that is doing exactly what we are building. However, there are companies working in related areas. Meshtastic enables off-grid communication but is open source and mainly designed for developers, requiring technical expertise to deploy. With EVE NET, users simply plug a USB-C cable into the EVE Box, and the network is ready to use. Quras allows developers to deploy applications on decentralised networks using blockchain, but we believe blockchain has become more centralised over time because transactions can be tracked. EVE NET allows developers to deploy applications without any fees, making the platform simpler and more cost-effective.
In India, Akash Network is the closest competitor in AI infrastructure. It provides a lower-cost alternative to cloud providers such as Google Cloud and Microsoft Azure by using multiple centralised servers in different locations, which it describes as decentralised infrastructure. We believe this is still based on centralised servers. Overall, no company currently offers the same combination of decentralised communication, compute, and storage that EVE NET provides.
Q. How have you funded the company so far?
A. We are currently bootstrapped and have not raised any external funding. The only capital we have used came from my previous startup, Velvet AI, which converted prompts into complete Android applications before tools like Cursor existed. Velvet AI was acquired for $50,000 through Acquired.com after about two years, and that funded our work on EVE NET.
Q. What is your marketing strategy for EVE NET?
A. At present, we are focused on the business-to-business B2B market. Since the company is only three months old, we have not needed a formal marketing strategy. Instead, I have been personally visiting universities and convincing them to deploy EVE routers within their existing ecosystems. We plan to develop a marketing strategy in the future, likely within the next month or two.
Q. Are you looking for channel partners or academic collaborations to expand your business?
A. We are not looking for channel or distribution partners for our hardware. Our EVE Boxes are mainly used by universities and construction companies, and demand has grown through customer referrals and word of mouth. We already collaborate with universities to develop and improve the technology, but these partnerships are focused on research rather than customer acquisition. Our main priority is building a strong developer ecosystem around the EVE SDK, which is expected to launch in the next two to three weeks. We want developers to adopt the EVE ecosystem as an alternative to platforms such as Firebase, Supabase, and other cloud backends, as we see this developer community as the key driver of our long-term growth.
Q. What are your current R&D priorities, and which applications are you targeting?
A. Our primary research and development (R&D) focus is securing data while it is distributed across the network. We are currently approaching law firms that want to use AI but cannot rely on cloud-based services because they handle highly sensitive legal data. With EVE AI, they can run AI workloads while keeping data decentralised, avoiding centralised cloud access. Beyond that, our research is focused on orchestration technology. Our long-term goal is to enable very large AI models, including models with up to one trillion parameters, to run across distributed mobile devices. Current orchestration technology is not yet powerful enough for this, but we are actively developing it.
Q. What are your expansion plans, and what opportunities do you see for EVE NET in the future?
A. Our immediate focus is to launch an SDK that allows developers to build and deploy applications on EVE NET while expanding the network by connecting more devices. Applications can run on both the internet and EVE NET, but on EVE NET, data is stored in a decentralised manner across connected devices instead of centralised servers.
Our next goal is to deploy a trillion-parameter AI model across decentralised mobile phones to deliver AI performance comparable to cloud-based services like ChatGPT. We also plan to support developers building AI applications, such as AI therapists and other AI wrappers, by providing compute and storage at a much lower cost using the unused resources of connected devices. Over the long term, we see strong opportunities in sectors such as banking and financial services, where decentralised data storage provides greater resilience against cyberattacks and outages by distributing replicated data across thousands of devices. Ultimately, our vision over the next 10 years is for EVE NET to become an alternative to today’s centralised internet, where a small number of companies control most of the infrastructure.
Q. What are your plans for future growth, and where are you investing today?
A. At present, most of our investment is going into R&D. Building the technology is our top priority right now. As we grow, we plan to invest heavily in marketing because convincing people to adopt a new off-grid network instead of relying on the internet will be a major challenge. So, while R&D is our current focus, marketing will become a key area of investment in the future.
