Atoma Network and Linera are integrating to enable scalable, private, and verifiable AI compute on the Linera blockchain.
Linera has partnered with Atoma Network to tap into Atoma’s decentralized AI inference network, bringing private and verifiable AI to the Linera ecosystem.
Linera is a novel blockchain designed for scalability and interoperability. Unlike traditional blockchains that rely on a single shared ledger, Linera is based on the concept of microchains which allow each user or application to process transactions independently in parallel chains, while sharing liquidity and data. Microchains also make Linera an ideal blockchain to interoperate with off-chain services, such as data storage and AI inference systems.
Atoma Network is an open-source decentralized AI cloud focused on privacy and verifiability. It aggregates global GPU resources to power AI applications, ensuring secure and permissionless access to AI inference. By leveraging trusted execution environments (TEEs) and a novel sampling consensus mechanism, Atoma enables developers to run AI models without exposing sensitive data and verifies the outputs.
Why does Web3 need decentralized AI?
AI is increasingly being adopted in Web3 applications for use cases such as agents, automation, data processing, and decision-making. Despite this, developers have to wrestle with the fact that these AI applications remain highly centralized.
This centralization exists at two key levels:
AI Model Providers – The most widely used AI models come from a handful of centralized companies like OpenAI, Anthropic, and Google. These companies control the model architectures, training data, and access policies, leaving users and developers with no transparency into how these models function.
Cloud Infrastructure – Even if developers use open-source AI models like those from Meta or DeepSeek, inference is still largely run on centralized cloud services like AWS, Azure, or GCP. These platforms own the compute resources required for AI inference, meaning they have full visibility into the data being processed.
Running AI inference through centralized providers means all requests are processed on third-party servers, giving cloud providers full access to user data. While AI performs better when it has access to more context, organizations must limit the data they share due to privacy concerns. This prevents them from fully leveraging AI’s potential, as using sensitive or proprietary data with centralized models risks exposure and loss of control.
AI models operate as black boxes, offering no transparency into their decision-making process or the integrity of their outputs. There is no way to verify whether an AI response has been altered, biased, or manipulated. Model providers and cloud platforms have full control over responses, often enforcing content restrictions through system prompts or model fine-tuning. This means AI models can be designed to refuse to answer or filter certain topics based on predefined guidelines (that the end user is not aware of). Additionally, providers can change model behavior at any time or revoke access without user control, leaving applications dependent on centralized decisions.
For Web3 projects, the importance of transparency, security, and availability are core pillars of the decentralized philosophy and a decentralized AI network can uphold these principles.
How does a decentralized AI network work with blockchains and dApps?
Traditional AI inferencing often relies on centralized servers, where a company such as OpenAI runs their own data centers. All AI requests run through those data centers with OpenAI having full control over the routing of the requests, visibility over the data being processed, and the integrity of the outputs.
Atoma instead partners with data centers across the world to intelligently (and randomly) distribute AI requests to suitable nodes on the network. To ensure full privacy, Atoma uses Trusted Execution Environments (TEEs) compatible with GPUs. Linera enables on-chain verification of Nvidia TEE attestations, reducing the need for trust.
For decentralized AI to be truly effective, it needs a blockchain capable of handling high-throughput AI requests without bottlenecks. Linera’s microchain model allows AI-powered applications to process transactions asynchronously, ensuring low-latency execution while preserving security guarantees. Atoma will provide Linera native access to decentralized AI compute and different open-source AI models.
How will Atoma and Linera work together?
Since Linera supports WebAssembly (Wasm) as its execution environment, any application compiled to Wasm can be adapted to run natively on Linera. Atoma has developed SDKs in Python, TypeScript, and soon Rust to facilitate direct communication between clients and Atoma’s decentralized AI compute network. These SDKs are designed to provide a Web2-like experience for developers to easily interact with Atoma’s infrastructure without deep blockchain expertise.
Atoma’s network operates through proxies, which serve as a middleware layer that matches compute buyers (clients) with compute providers (nodes). These proxies track node performance, uptime, and hardware specifications, ensuring efficient allocation of AI compute resources. While proxies authenticate client requests, all sensitive operations—such as encryption, decryption, and client signature verification—are handled securely within Atoma’s SDKs. With these SDKs as Wasm-compatible, Linera developers will be able to import an Atoma client library, authenticate AI compute requests, and allow Linera nodes to seamlessly access Atoma’s AI inference layer within their applications.
Beyond direct AI compute integration, Linera nodes can verify public and private web data before passing it to Atoma’s AI inference engine. This is particularly relevant for autonomous AI workflows, such as AI agents managing on-chain transactions or executing tasks based on verified off-chain inputs.
To ensure a simple authentication process, Atoma proxies provide a Web2-like authentication method using API keys. This approach works out of the box in Linera thanks to the platform’s native support for Web2 queries. In the future, Linera validators will integrate with the Sui blockchain and be able to verify Atoma’s responses directly
Whenever an application requests AI compute from Atoma, it generates a random nonce, leveraging Sui’s on-chain randomness in the interim. Once Linera's native on-chain randomness becomes available, it will be used. This nonce uniquely identifies each transaction request and is verified by Linera nodes. The Atoma proxy will only accept the request if a supermajority of Linera nodes have signed the associated transaction, ensuring decentralized verification while maintaining efficiency.
Additionally, Atoma’s AI compute can be used to enhance block production on Linera, introducing AI-powered optimizations for network efficiency and validator operations. By building a native Linera Atoma SDK that incorporates these authentication and compute request mechanisms, Atoma can integrate into Linera’s ecosystem while maintaining its core principles of verifiability, privacy, and decentralization.
How can you start using Atoma’s AI inference on Linera?
Developers can start integrating Atoma’s AI on Linera in just a few steps:
1. Install the Atoma SDK
Atoma provides SDKs in Python, TypeScript, and soon Rust designed for Web2-like simplicity. Navigate to the documentation for more information.
2. Authenticate AI Compute Requests
Applications authenticate using an API key-based system (temporary Web2-like authentication).
3. Submit an Inference Request
Developers send AI inference requests via the SDK, specifying the model, data input, and any required parameters.
Atoma’s proxy layer routes the request to a node based on performance, hardware availability, and network conditions.
4. AI Processing & Verifiable Output
The inference runs inside trusted execution environments (TEEs), ensuring data privacy and output integrity.
Results are cryptographically verified before being returned to the requesting application.
5. Use AI in a Linera-based dApp
AI-generated outputs can be used for smart contract automation, AI-driven governance, DeFi analytics, on-chain agents and more.
Developers can fetch AI results on-chain or use them off-chain for additional processing.
Try the Atoma AI Demo on Linera
To see Atoma’s AI inference in action on Linera, check out the Atoma-Linera demo on GitHub.
The repository provides example code, setup instructions, and a walkthrough of making AI requests on Linera.