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of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time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of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Basilica (Bittensor Subnet 39) is a decentralized marketplace for high-performance GPU computing power, built on the Bittensor network. It connects GPU providers (known as miners) with users who need to run intensive AI workloads (like training or inference tasks) in a trustless environment. The platform ensures that GPU providers can offer compute resources securely while validators (special nodes on the network) verify the quality and reliability of the computation through cryptographic methods. In essence, Basilica turns distributed GPUs into a global supercomputer accessible on-demand, with blockchain-based incentives to guarantee honest and efficient service delivery. It prioritizes performance and reliability of compute over simple price bidding: a performance-first approach where nodes are rewarded for delivering consistent, high-quality results rather than just the lowest cost. To enforce this, Basilica uses smart-contract-based collateral mechanisms – miners stake collateral that can be slashed for bad service – ensuring that tasks are completed correctly and on time. This creates a self-regulating marketplace where miners compete to provide reliable GPU compute and validators rigorously evaluate their outputs, with all interactions recorded on Bittensor’s blockchain.
Basilica’s target users are AI researchers, developers, and organizations that require scalable and trustworthy GPU computing for machine learning tasks. Instead of relying on centralized cloud GPU providers, they can “summon compute” from Basilica’s decentralized network. The Bittensor substrate blockchain underpins Basilica’s economy – validators assign weights to miners based on the value of their contributions, which in turn governs the mining rewards in TAO (the Bittensor token). In summary, Basilica offers a way to rent out or utilize GPU power without trusting a central entity, because the network itself handles verification of hardware and results, and rewards only those providers who perform well. This is a critical piece of the broader Covenant AI ecosystem, serving as the infrastructure layer alongside Subnet 3 Templar (focused on model pre-training) and Subnet 81 Grail (focused on fine-tuning), to enable end-to-end decentralized AI development.
Basilica (Bittensor Subnet 39) is a decentralized marketplace for high-performance GPU computing power, built on the Bittensor network. It connects GPU providers (known as miners) with users who need to run intensive AI workloads (like training or inference tasks) in a trustless environment. The platform ensures that GPU providers can offer compute resources securely while validators (special nodes on the network) verify the quality and reliability of the computation through cryptographic methods. In essence, Basilica turns distributed GPUs into a global supercomputer accessible on-demand, with blockchain-based incentives to guarantee honest and efficient service delivery. It prioritizes performance and reliability of compute over simple price bidding: a performance-first approach where nodes are rewarded for delivering consistent, high-quality results rather than just the lowest cost. To enforce this, Basilica uses smart-contract-based collateral mechanisms – miners stake collateral that can be slashed for bad service – ensuring that tasks are completed correctly and on time. This creates a self-regulating marketplace where miners compete to provide reliable GPU compute and validators rigorously evaluate their outputs, with all interactions recorded on Bittensor’s blockchain.
Basilica’s target users are AI researchers, developers, and organizations that require scalable and trustworthy GPU computing for machine learning tasks. Instead of relying on centralized cloud GPU providers, they can “summon compute” from Basilica’s decentralized network. The Bittensor substrate blockchain underpins Basilica’s economy – validators assign weights to miners based on the value of their contributions, which in turn governs the mining rewards in TAO (the Bittensor token). In summary, Basilica offers a way to rent out or utilize GPU power without trusting a central entity, because the network itself handles verification of hardware and results, and rewards only those providers who perform well. This is a critical piece of the broader Covenant AI ecosystem, serving as the infrastructure layer alongside Subnet 3 Templar (focused on model pre-training) and Subnet 81 Grail (focused on fine-tuning), to enable end-to-end decentralized AI development.
Key Features: Basilica achieves its trustless GPU marketplace via several core mechanisms:
Hardware Verification – Every participating GPU miner must run a secure binary that validates and profiles their GPU hardware, proving its specs and capabilities to the network. This prevents miners from misreporting their GPU type or performance.
Remote Validation – Basilica’s validators can remotely verify computations and hardware by establishing secure SSH connections to miners. They may run test tasks or inspections on a miner’s machine to ensure that jobs are executed correctly and that the system matches the reported specifications.
Bittensor Integration – The subnet is natively integrated into Bittensor’s consensus mechanism. Basilica validators participate in Bittensor’s metagraph (the network’s consensus graph) by assigning weights to miners based on successful task execution and reliability. This weight directly influences how much reward each miner earns, aligning miners’ incentives with providing accurate and useful compute service.
Fleet Management – Basilica miners can manage entire fleets of GPUs. A miner node orchestrates multiple executors (worker processes on GPU machines) and handles assignment of incoming compute jobs across these resources efficiently. The system tracks which GPU executors are free or busy and distributes tasks to optimize utilization.
API Gateway – Basilica provides a smart HTTP API gateway that allows external clients to submit workloads to the network in a user-friendly way. This gateway load-balances requests across the validator network and offers features like authentication, caching of results, rate limiting, and aggregation of responses. It essentially serves as the front-end for developers to access the decentralized compute network without needing to interact with the blockchain directly.
These features collectively enable secure, verifiable, and efficient GPU computing on a decentralized network. For example, when a user submits an AI training job via the Basilica API, the request is routed to a miner’s Axon (Bittensor server interface) which then schedules the job on one of its GPUs (the executor). Validators might randomly spot-check the computation or require cryptographic proofs of execution. Once the job is done, results return through the gateway to the user. The blockchain records the performance: miners who completed tasks correctly and fast get high weight (and thus more TAO reward), whereas any miner that tried to cheat or produced incorrect results would be detected by validators (due to verification checks) and penalized by losing weight or collateral. This design ensures quality of service in a trustless setting.
Basilica’s product is a decentralized compute network – essentially a software stack that anyone can run to either provide GPU power or request GPU-based computation. The build consists of several interconnected components, mostly implemented in Rust (over 94% of the codebase). Technically, Basilica is delivered as an open-source protocol and software client that runs on the Bittensor substrate network (with Subnet ID 39 for mainnet). Users install the Basilica software (for example via a one-line installer script as shown on the website) and can run it in different modes depending on their role:
Miner Node: The miner is the component a GPU provider runs. It registers on the Bittensor network as a Basilica miner and manages one or more GPUs (executors). The miner software handles task scheduling, communicates with validators, and exposes a Bittensor Axon server to receive compute requests. Miners also include logic to run the hardware verification binaries and to accept remote checks from validators. In practice, a miner will coordinate multiple GPU machines: each physical machine (or container) runs an Executor agent that reports to the miner. The miner allocates incoming jobs to these executors, monitors their status, and returns results. It’s also responsible for staking any required collateral and following the network’s protocol rules (e.g., responding within timeouts).
Executor: This is the lightweight agent that runs on each GPU machine under a miner’s control. Executors listen for tasks (such as a chunk of a training job or an inference request) from the miner and run them in a secure containerized environment. They also gather system metrics and ensure the task is isolated. The executor is essentially the worker that actually performs the computation on the GPU, while the miner is the overseer. Executors must run a locked-down environment for security – possibly using Docker or similar – as hinted by “container management” in the docs. They report back results and any execution proof data to the miner.
Validator Node: The validator is the backbone of trust in Basilica. Validators are special nodes (typically run by distinct participants, potentially requiring significant reputation or stake) that verify miners’ claims and performance. A validator maintains a profile of each miner’s hardware (e.g., the GPU model, CUDA cores, memory, compute benchmarks) and continuously scores miner performance during network operations. Validators may issue challenge tasks or use the remote execution feature to test that a miner’s GPU is genuine and that it produces correct outputs. They also validate results of real jobs when possible. Importantly, each validator then uses Bittensor’s consensus mechanism to assign a weight (a numeric score) to every miner based on these evaluations. This weight is broadcast on-chain. Miners with higher weights get a larger portion of the next block’s TAO issuance (reward), whereas those with low or zero weight earn little, aligning economic reward with trustworthiness and performance. Validators themselves also earn rewards (they are like “verifiers” in the system) and must be reliable. In Basilica’s design, validators form a distributed auditing layer ensuring the compute marketplace remains honest.
Basilica API Gateway: This is a service layer that the Basilica team provides (or that can be run by anyone) to make it easier for end-users to access the compute network. Instead of directly interacting with miners via the Bittensor protocol, a user can send HTTP requests to the gateway. The gateway will handle finding available miners/validators, load-balancing the requests among the network, aggregating results if needed, and returning the output to the user. It also includes authentication (to ensure only authorized or paying users use the compute), caching (so repeated queries might be served faster if results can be reused), and rate limiting (to prevent abuse or overload). Essentially, the gateway turns the decentralized network into a cloud-like API service for developers – you could imagine an AI researcher calling the Basilica API to get, say, 100 GPU hours of training on their model, without worrying about which specific miner does the work. The gateway would orchestrate that across the subnet’s miners behind the scenes.
Common Libraries & Protocols: Basilica’s codebase includes common modules (“Basilica Common”) that provide shared functionality like cryptography (for signing and verifying messages, managing keys), SSH management (for the remote validation sessions), data storage (perhaps storing hardware profiles, telemetry), and configuration management. It also defines a Protocol layer using gRPC/protobuf for all the internal communications between executors, miners, validators, and the API gateway. By using gRPC, the system ensures efficient and typed communication across the distributed components. All participants speak the same protocol for tasks such as job assignment, result reporting, heartbeats, etc., which makes the network cohesive.
To summarize the build: Basilica is essentially a distributed system comprising client software that miners and validators run to form the network, plus auxiliary services (like the API gateway) to make it accessible. The product isn’t a singular hardware device or a SaaS platform in the traditional sense, but rather an open protocol and network. From a user’s perspective, the “product” is the ability to submit AI compute jobs to a decentralized network of GPUs and get results with assurances of trust and quality. From a miner’s perspective, the product is a software that lets them monetize their GPU hardware by contributing to this network. All of it is enabled by fairly complex under-the-hood architecture coordinating hardware verification, job scheduling, and blockchain consensus.
Key Features: Basilica achieves its trustless GPU marketplace via several core mechanisms:
Hardware Verification – Every participating GPU miner must run a secure binary that validates and profiles their GPU hardware, proving its specs and capabilities to the network. This prevents miners from misreporting their GPU type or performance.
Remote Validation – Basilica’s validators can remotely verify computations and hardware by establishing secure SSH connections to miners. They may run test tasks or inspections on a miner’s machine to ensure that jobs are executed correctly and that the system matches the reported specifications.
Bittensor Integration – The subnet is natively integrated into Bittensor’s consensus mechanism. Basilica validators participate in Bittensor’s metagraph (the network’s consensus graph) by assigning weights to miners based on successful task execution and reliability. This weight directly influences how much reward each miner earns, aligning miners’ incentives with providing accurate and useful compute service.
Fleet Management – Basilica miners can manage entire fleets of GPUs. A miner node orchestrates multiple executors (worker processes on GPU machines) and handles assignment of incoming compute jobs across these resources efficiently. The system tracks which GPU executors are free or busy and distributes tasks to optimize utilization.
API Gateway – Basilica provides a smart HTTP API gateway that allows external clients to submit workloads to the network in a user-friendly way. This gateway load-balances requests across the validator network and offers features like authentication, caching of results, rate limiting, and aggregation of responses. It essentially serves as the front-end for developers to access the decentralized compute network without needing to interact with the blockchain directly.
These features collectively enable secure, verifiable, and efficient GPU computing on a decentralized network. For example, when a user submits an AI training job via the Basilica API, the request is routed to a miner’s Axon (Bittensor server interface) which then schedules the job on one of its GPUs (the executor). Validators might randomly spot-check the computation or require cryptographic proofs of execution. Once the job is done, results return through the gateway to the user. The blockchain records the performance: miners who completed tasks correctly and fast get high weight (and thus more TAO reward), whereas any miner that tried to cheat or produced incorrect results would be detected by validators (due to verification checks) and penalized by losing weight or collateral. This design ensures quality of service in a trustless setting.
Basilica’s product is a decentralized compute network – essentially a software stack that anyone can run to either provide GPU power or request GPU-based computation. The build consists of several interconnected components, mostly implemented in Rust (over 94% of the codebase). Technically, Basilica is delivered as an open-source protocol and software client that runs on the Bittensor substrate network (with Subnet ID 39 for mainnet). Users install the Basilica software (for example via a one-line installer script as shown on the website) and can run it in different modes depending on their role:
Miner Node: The miner is the component a GPU provider runs. It registers on the Bittensor network as a Basilica miner and manages one or more GPUs (executors). The miner software handles task scheduling, communicates with validators, and exposes a Bittensor Axon server to receive compute requests. Miners also include logic to run the hardware verification binaries and to accept remote checks from validators. In practice, a miner will coordinate multiple GPU machines: each physical machine (or container) runs an Executor agent that reports to the miner. The miner allocates incoming jobs to these executors, monitors their status, and returns results. It’s also responsible for staking any required collateral and following the network’s protocol rules (e.g., responding within timeouts).
Executor: This is the lightweight agent that runs on each GPU machine under a miner’s control. Executors listen for tasks (such as a chunk of a training job or an inference request) from the miner and run them in a secure containerized environment. They also gather system metrics and ensure the task is isolated. The executor is essentially the worker that actually performs the computation on the GPU, while the miner is the overseer. Executors must run a locked-down environment for security – possibly using Docker or similar – as hinted by “container management” in the docs. They report back results and any execution proof data to the miner.
Validator Node: The validator is the backbone of trust in Basilica. Validators are special nodes (typically run by distinct participants, potentially requiring significant reputation or stake) that verify miners’ claims and performance. A validator maintains a profile of each miner’s hardware (e.g., the GPU model, CUDA cores, memory, compute benchmarks) and continuously scores miner performance during network operations. Validators may issue challenge tasks or use the remote execution feature to test that a miner’s GPU is genuine and that it produces correct outputs. They also validate results of real jobs when possible. Importantly, each validator then uses Bittensor’s consensus mechanism to assign a weight (a numeric score) to every miner based on these evaluations. This weight is broadcast on-chain. Miners with higher weights get a larger portion of the next block’s TAO issuance (reward), whereas those with low or zero weight earn little, aligning economic reward with trustworthiness and performance. Validators themselves also earn rewards (they are like “verifiers” in the system) and must be reliable. In Basilica’s design, validators form a distributed auditing layer ensuring the compute marketplace remains honest.
Basilica API Gateway: This is a service layer that the Basilica team provides (or that can be run by anyone) to make it easier for end-users to access the compute network. Instead of directly interacting with miners via the Bittensor protocol, a user can send HTTP requests to the gateway. The gateway will handle finding available miners/validators, load-balancing the requests among the network, aggregating results if needed, and returning the output to the user. It also includes authentication (to ensure only authorized or paying users use the compute), caching (so repeated queries might be served faster if results can be reused), and rate limiting (to prevent abuse or overload). Essentially, the gateway turns the decentralized network into a cloud-like API service for developers – you could imagine an AI researcher calling the Basilica API to get, say, 100 GPU hours of training on their model, without worrying about which specific miner does the work. The gateway would orchestrate that across the subnet’s miners behind the scenes.
Common Libraries & Protocols: Basilica’s codebase includes common modules (“Basilica Common”) that provide shared functionality like cryptography (for signing and verifying messages, managing keys), SSH management (for the remote validation sessions), data storage (perhaps storing hardware profiles, telemetry), and configuration management. It also defines a Protocol layer using gRPC/protobuf for all the internal communications between executors, miners, validators, and the API gateway. By using gRPC, the system ensures efficient and typed communication across the distributed components. All participants speak the same protocol for tasks such as job assignment, result reporting, heartbeats, etc., which makes the network cohesive.
To summarize the build: Basilica is essentially a distributed system comprising client software that miners and validators run to form the network, plus auxiliary services (like the API gateway) to make it accessible. The product isn’t a singular hardware device or a SaaS platform in the traditional sense, but rather an open protocol and network. From a user’s perspective, the “product” is the ability to submit AI compute jobs to a decentralized network of GPUs and get results with assurances of trust and quality. From a miner’s perspective, the product is a software that lets them monetize their GPU hardware by contributing to this network. All of it is enabled by fairly complex under-the-hood architecture coordinating hardware verification, job scheduling, and blockchain consensus.
Basilica is developed by the Covenant AI team (also referred to as Templar AI, reflecting their Subnet 3 project). It’s a small, highly specialized team of engineers and researchers at the forefront of decentralized AI and blockchain. Key members of the team include:
Samuel Dare – Project lead for Basilica. Sam is described as a blockchain veteran and the visionary behind the “Sacred Compute” concept of Basilica. He guides the overall architecture and ensures the product meets both blockchain and AI requirements. In the academic context, Sam is a co-author of the research on permissionless distributed training that underpins parts of this project, indicating his deep involvement in both the theoretical and practical aspects of the network.
Evangelos (“Evan”) Pappas – Co-founder of Covenant AI and core developer. Evan often represents the project publicly (for instance, on social media he evangelizes Bittensor and Covenant’s mission). He has a strong vision for decentralized AI’s future – even stating that dedicating his life to building on Bittensor would be worthwhile. Evan’s role spans from strategic decisions about the ecosystem (Templar, Basilica, Grail coordination) to hands-on development. He is also listed as a co-author in the team’s research paper, confirming his technical contributions.
Joel Lidin – AI Researcher and Engineer. Joel’s focus is on the machine learning side of things – ensuring that the incentive mechanisms (like the Gauntlet system for evaluating contributions) are grounded in solid ML principles. Coming from an AI research background (and also a co-author on the LLM training incentive paper), he contributes to the design of algorithms that validators use to score miners, as well as optimizations for distributed training. Joel is deeply involved in both Basilica and the upcoming Grail subnet (his online bio humorously calls him a “Summoner of Divine Computation” for Basilica/Grail).
Amir Sarfi – AI Researcher and Engineer. Amir, like Joel, is an expert in machine learning and distributed systems. He is also among the authors of the academic paper on decentralized training. Amir’s expertise likely lies in the training dynamics and mathematical modeling of the incentive and verification systems. He helps design how Basilica’s validators can quickly yet accurately verify work (for example, devising efficient proofs or checks for GPU computations) and contributes to the overall protocol development.
These four individuals are all affiliated with Templar AI (tplr.ai) and form the core development team of Basilica. The project is highly interdisciplinary, so the team blends skills in blockchain (Sam, Evan) and AI research (Joel, Amir) to cover all aspects. Additionally, the team collaborates with prominent figures in the space: for example, Eugene Belilovsky (a researcher from MILA/Concordia University) and Jacob Steeves (co-founder of the Opentensor Foundation, which created Bittensor) have worked with them on research and possibly advise on the project. While not part of the day-to-day Basilica dev team, their involvement (e.g., as co-authors on research) signals strong connections to academia and the broader Bittensor community.
It’s worth noting that “One Covenant, Many Orders” is the motto on the Covenant AI website – Templar, Basilica, and Grail can be seen as three “orders” or sub-projects under one team. So the team above is effectively the same group driving all these subnets, with Basilica being Subnet 39 under their stewardship. They operate out of locations like San Francisco (per GitHub) and are active on Discord/X under the handles @tplr_ai (Templar), @basilic_ai (Basilica), and @grail_ai. This close-knit team has rapidly delivered innovations like the first permissionless 70B parameter model training (via Templar) and now the decentralized compute network (Basilica), showing a clear track record in the DeAI (Decentralized AI) space.
Basilica is developed by the Covenant AI team (also referred to as Templar AI, reflecting their Subnet 3 project). It’s a small, highly specialized team of engineers and researchers at the forefront of decentralized AI and blockchain. Key members of the team include:
Samuel Dare – Project lead for Basilica. Sam is described as a blockchain veteran and the visionary behind the “Sacred Compute” concept of Basilica. He guides the overall architecture and ensures the product meets both blockchain and AI requirements. In the academic context, Sam is a co-author of the research on permissionless distributed training that underpins parts of this project, indicating his deep involvement in both the theoretical and practical aspects of the network.
Evangelos (“Evan”) Pappas – Co-founder of Covenant AI and core developer. Evan often represents the project publicly (for instance, on social media he evangelizes Bittensor and Covenant’s mission). He has a strong vision for decentralized AI’s future – even stating that dedicating his life to building on Bittensor would be worthwhile. Evan’s role spans from strategic decisions about the ecosystem (Templar, Basilica, Grail coordination) to hands-on development. He is also listed as a co-author in the team’s research paper, confirming his technical contributions.
Joel Lidin – AI Researcher and Engineer. Joel’s focus is on the machine learning side of things – ensuring that the incentive mechanisms (like the Gauntlet system for evaluating contributions) are grounded in solid ML principles. Coming from an AI research background (and also a co-author on the LLM training incentive paper), he contributes to the design of algorithms that validators use to score miners, as well as optimizations for distributed training. Joel is deeply involved in both Basilica and the upcoming Grail subnet (his online bio humorously calls him a “Summoner of Divine Computation” for Basilica/Grail).
Amir Sarfi – AI Researcher and Engineer. Amir, like Joel, is an expert in machine learning and distributed systems. He is also among the authors of the academic paper on decentralized training. Amir’s expertise likely lies in the training dynamics and mathematical modeling of the incentive and verification systems. He helps design how Basilica’s validators can quickly yet accurately verify work (for example, devising efficient proofs or checks for GPU computations) and contributes to the overall protocol development.
These four individuals are all affiliated with Templar AI (tplr.ai) and form the core development team of Basilica. The project is highly interdisciplinary, so the team blends skills in blockchain (Sam, Evan) and AI research (Joel, Amir) to cover all aspects. Additionally, the team collaborates with prominent figures in the space: for example, Eugene Belilovsky (a researcher from MILA/Concordia University) and Jacob Steeves (co-founder of the Opentensor Foundation, which created Bittensor) have worked with them on research and possibly advise on the project. While not part of the day-to-day Basilica dev team, their involvement (e.g., as co-authors on research) signals strong connections to academia and the broader Bittensor community.
It’s worth noting that “One Covenant, Many Orders” is the motto on the Covenant AI website – Templar, Basilica, and Grail can be seen as three “orders” or sub-projects under one team. So the team above is effectively the same group driving all these subnets, with Basilica being Subnet 39 under their stewardship. They operate out of locations like San Francisco (per GitHub) and are active on Discord/X under the handles @tplr_ai (Templar), @basilic_ai (Basilica), and @grail_ai. This close-knit team has rapidly delivered innovations like the first permissionless 70B parameter model training (via Templar) and now the decentralized compute network (Basilica), showing a clear track record in the DeAI (Decentralized AI) space.
Basilica launched on Bittensor mainnet in mid-2025 and has been rapidly iterating since. For instance, the team has pushed frequent software updates (the Basilica CLI was at version 0.3.3 as of September 19, 2025, indicating active development). The short-term roadmap is focused on stabilizing and scaling the network – ensuring that the core features (hardware verification, task scheduling, etc.) work reliably as more miners and users join. In the weeks following launch, Basilica already achieved a “heartbeat” (the network became alive and operational) and demonstrated basic functionality with initial GPU providers. The developers are likely now working on performance optimizations (to handle larger workloads and more concurrent tasks) and refining the validation algorithms so that even as the network grows, it can accurately score miners and catch any faulty behavior.
A priority in the roadmap is to grow the supply of GPU resources on the network. This means outreach to get more miners onboarded – potentially by lowering the barrier to entry (easy installation like the one-line install script) and proving that miners can earn good rewards. On the demand side, we can expect efforts to integrate Basilica with user-friendly interfaces or ML frameworks. For example, a future update might allow an AI developer to plug Basilica’s compute directly into a Python ML library or a cloud workflow. The Basilica API gateway will be an important part of this; going forward, the team will likely expand its capabilities (more robust scheduling of jobs across multiple miners, billing and payment integration for users, etc.). Essentially, moving from the current “early adopter” stage to a more polished product experience is on the roadmap.
In terms of features, additional verification and security enhancements are anticipated. The team has hinted at exploring advanced methods like zero-knowledge proofs for ML outputs or trustless enclave execution, although they have been careful not to prematurely confirm specific features. (When asked about certain potential features, team members responded that they could not confirm or deny if those are part of the roadmap, suggesting they do have plans but aren’t ready to announce them.) This means we might see Basilica incorporate cutting-edge techniques to further ensure that results returned by miners are correct without needing to fully re-compute them – a challenging but game-changing capability for a decentralized compute network.
Another aspect of the roadmap is deeper integration with the rest of Covenant AI’s ecosystem. Since Basilica provides the raw compute layer, it will underpin Templar and Grail. We can expect coordinated upgrades: for example, as Templar (Subnet 3) launches larger decentralized model training runs, Basilica may need to support those by allocating lots of GPUs and handling gradient aggregation workloads. The roadmap likely includes supporting new workload types (e.g., not just standard AI model training, but perhaps also reinforcement learning loops for Grail, or other compute-intensive AI tasks). Each of these might require tweaks to how tasks are scheduled or validated. Thus, Basilica’s development will closely follow the needs of those AI workloads.
On the community/governance side, because Basilica operates on Bittensor, any major changes (like protocol upgrades) might go through the Bittensor on-chain governance process. The team will roadmap proposals for improvements (BIPs or “BITIPs” for Bittensor Improvement Proposals) as needed. Future milestones could include achieving a certain market cap or usage metric – interestingly, within a short time of launch, Basilica’s subnet token value grew significantly, contributing to Bittensor’s subnet market cap surge. Maintaining that growth means executing on the technical roadmap and demonstrating real-world utility.
Long-term, the vision for Basilica is to become a global, decentralized supercomputer. According to statements associated with the project, they envision harnessing the untapped GPU power of everyday devices – not just data center GPUs, but also gaming PCs and even smartphones in the future. Achieving this will likely extend beyond 2025: it involves making the software lightweight and secure enough to run on various hardware, and creating incentive models for smaller contributors. The roadmap, therefore, includes mass decentralization and democratization of AI compute. Over time, we might see features like mobile miners or browser-based miners if technology permits. The endgame is a network where anyone can contribute computing power and anyone can access massive compute on demand, truly realizing AI that is “decentralized, democratized, and accessible to everyone”.
In summary, Basilica’s roadmap can be thought of in phases: (1) harden and scale the current network (improve reliability, throughput, and miner/validator participation), (2) integrate with user platforms and Covenant’s other subnets to drive usage, (3) introduce advanced verification and possibly economic features like more sophisticated collateral or pricing models (ensuring the marketplace remains fair and robust), and (4) expand the network’s reach to more device types and a broader community. While specific feature rollouts haven’t all been publicized, the ongoing updates and the stated vision give a clear picture that Basilica is just at the beginning of its journey. Every update brings it a step closer to that “holy grail” of a decentralized AI infrastructure where compute power flows as freely and securely as information on the internet. The team’s commitment to continuous improvement is evident – as one community member noted, Bittensor subnets like Basilica can “continuously improve without limit” as live open marketplaces, so we can expect Basilica to evolve rapidly with no rigid end-state. The coming months and years should see Basilica transform from a promising Subnet into a critical piece of the AI compute landscape, guided by the ambitious roadmap laid out by its creators.
Basilica launched on Bittensor mainnet in mid-2025 and has been rapidly iterating since. For instance, the team has pushed frequent software updates (the Basilica CLI was at version 0.3.3 as of September 19, 2025, indicating active development). The short-term roadmap is focused on stabilizing and scaling the network – ensuring that the core features (hardware verification, task scheduling, etc.) work reliably as more miners and users join. In the weeks following launch, Basilica already achieved a “heartbeat” (the network became alive and operational) and demonstrated basic functionality with initial GPU providers. The developers are likely now working on performance optimizations (to handle larger workloads and more concurrent tasks) and refining the validation algorithms so that even as the network grows, it can accurately score miners and catch any faulty behavior.
A priority in the roadmap is to grow the supply of GPU resources on the network. This means outreach to get more miners onboarded – potentially by lowering the barrier to entry (easy installation like the one-line install script) and proving that miners can earn good rewards. On the demand side, we can expect efforts to integrate Basilica with user-friendly interfaces or ML frameworks. For example, a future update might allow an AI developer to plug Basilica’s compute directly into a Python ML library or a cloud workflow. The Basilica API gateway will be an important part of this; going forward, the team will likely expand its capabilities (more robust scheduling of jobs across multiple miners, billing and payment integration for users, etc.). Essentially, moving from the current “early adopter” stage to a more polished product experience is on the roadmap.
In terms of features, additional verification and security enhancements are anticipated. The team has hinted at exploring advanced methods like zero-knowledge proofs for ML outputs or trustless enclave execution, although they have been careful not to prematurely confirm specific features. (When asked about certain potential features, team members responded that they could not confirm or deny if those are part of the roadmap, suggesting they do have plans but aren’t ready to announce them.) This means we might see Basilica incorporate cutting-edge techniques to further ensure that results returned by miners are correct without needing to fully re-compute them – a challenging but game-changing capability for a decentralized compute network.
Another aspect of the roadmap is deeper integration with the rest of Covenant AI’s ecosystem. Since Basilica provides the raw compute layer, it will underpin Templar and Grail. We can expect coordinated upgrades: for example, as Templar (Subnet 3) launches larger decentralized model training runs, Basilica may need to support those by allocating lots of GPUs and handling gradient aggregation workloads. The roadmap likely includes supporting new workload types (e.g., not just standard AI model training, but perhaps also reinforcement learning loops for Grail, or other compute-intensive AI tasks). Each of these might require tweaks to how tasks are scheduled or validated. Thus, Basilica’s development will closely follow the needs of those AI workloads.
On the community/governance side, because Basilica operates on Bittensor, any major changes (like protocol upgrades) might go through the Bittensor on-chain governance process. The team will roadmap proposals for improvements (BIPs or “BITIPs” for Bittensor Improvement Proposals) as needed. Future milestones could include achieving a certain market cap or usage metric – interestingly, within a short time of launch, Basilica’s subnet token value grew significantly, contributing to Bittensor’s subnet market cap surge. Maintaining that growth means executing on the technical roadmap and demonstrating real-world utility.
Long-term, the vision for Basilica is to become a global, decentralized supercomputer. According to statements associated with the project, they envision harnessing the untapped GPU power of everyday devices – not just data center GPUs, but also gaming PCs and even smartphones in the future. Achieving this will likely extend beyond 2025: it involves making the software lightweight and secure enough to run on various hardware, and creating incentive models for smaller contributors. The roadmap, therefore, includes mass decentralization and democratization of AI compute. Over time, we might see features like mobile miners or browser-based miners if technology permits. The endgame is a network where anyone can contribute computing power and anyone can access massive compute on demand, truly realizing AI that is “decentralized, democratized, and accessible to everyone”.
In summary, Basilica’s roadmap can be thought of in phases: (1) harden and scale the current network (improve reliability, throughput, and miner/validator participation), (2) integrate with user platforms and Covenant’s other subnets to drive usage, (3) introduce advanced verification and possibly economic features like more sophisticated collateral or pricing models (ensuring the marketplace remains fair and robust), and (4) expand the network’s reach to more device types and a broader community. While specific feature rollouts haven’t all been publicized, the ongoing updates and the stated vision give a clear picture that Basilica is just at the beginning of its journey. Every update brings it a step closer to that “holy grail” of a decentralized AI infrastructure where compute power flows as freely and securely as information on the internet. The team’s commitment to continuous improvement is evident – as one community member noted, Bittensor subnets like Basilica can “continuously improve without limit” as live open marketplaces, so we can expect Basilica to evolve rapidly with no rigid end-state. The coming months and years should see Basilica transform from a promising Subnet into a critical piece of the AI compute landscape, guided by the ambitious roadmap laid out by its creators.
A big thank you to Tao Stats for producing these insightful videos in the Novelty Search series. We appreciate the opportunity to dive deep into the groundbreaking work being done by Subnets within Bittensor! Check out some of their other videos HERE.
Recorded August 2025: This Tao Stats Novelty Search session wrangles a lively panel to unveil Covenant’s three-part push across Bittensor—Templar (pre-training), Basilica (compute), and Grail (post-training/RL). The team spotlights a big research leap: Sparse-LoCo, a decentralized training optimizer that combines top-k compression with 2-bit quantization to slash communication while improving accuracy, enabling a permissionless 70B-parameter run. Basilica is positioned as a compute network that will evolve beyond “rentals” into value-added services like verifiable inference and hardware-efficiency tricks to cut the “Jensen tax.” Grail targets single- then multi-turn RL, plus a fast hidden-state “fingerprint” to verify miners’ outputs and model usage. Together—with a coming rebrand to Covenant.ai—the trio aims to turn open research into production pipelines while keeping incentives aligned and results shareable.
TGIF! Join @DistStateAndMe for our Community X Space TODAY
Updates from across the @covenant_ai ecosystem ( @tplr_ai + @basilic_ai + @grail_ai ) and a casual chat with the Bittensor community
Come hang out, ask questions, share ideas
🔗 https://x.com/i/spaces/1ZkKzZWnVRRKv
come and get it anon @basilic_ai
They say the first 0.1 Tao is always the hardest ...
@basilic_ai has revenue!
We plan to ramp this up significantly in the coming weeks.
Rentals are just the beginning.
https://taostats.io/account/5H14o4Xoe5vquvD9fZGYWUS5T6KnzgtCBqffQEmcJpRRzpsf/transfers
Decentralised compute networks are hard.
Miners , as lovely as they are , are rational , profit maximising, entities.
They are motivated to provide interruptible instances , which get yanked by providers.
1/n
ὁ ζητῶν τὸν GPU-Compute εὑρίσκει εἰς τὴν http://basilica.ai
Things are accelerating @basilic_ai .
We are at a 90% utilisation rate, and work is in the background to enable fleets of GPUs.
Should land this week or next.
Renting a B200*8 is as simple as "basilica up"
What are you waiting for anon?
https://www.basilica.ai/
