Atoma Network primary sources
Reviewed: 2026-05-14 UTC
Public website / current positioning
- https://www.atoma.network/
- Currently redirects to
https://atoma.ai/. - Used to confirm the current public-facing framing: secure AI infrastructure, confidential computing, attestation, GPU-backed secure nodes, and enterprise-oriented workload protection.
- Important because it shows a newer commercialization layer that is more enterprise-security-forward than the older protocol docs and repositories.
- Currently redirects to
Core protocol introduction
- https://atoma.gitbook.io/atoma-docs/introduction/atoma-basics.md
- Best concise protocol-level overview in this pass.
- Used to confirm that Atoma presents itself as a decentralized, scalable, permissionless AI-computing protocol built around execution nodes, Sampling Consensus, and TEEs.
Compute-market and routing design
- https://atoma.gitbook.io/atoma-docs/introduction/compute-layer.md
- Best source for the network and marketplace architecture.
- Used to confirm the decentralized execution-node pool, intelligent request routing based on cost / uptime / privacy / latency / hardware / workload, node reputation, collateral, and the docs’ own framing of Atoma as a free market for AI compute.
- Also useful for seeing Atoma distinguish itself from generic DePIN by making workload specialization, routing logic, and performance / security tuning explicit.
Verification and privacy design
- https://atoma.gitbook.io/atoma-docs/introduction/trust-and-privacy.md
- Best source for Atoma’s core mechanism claims.
- Used to confirm the protocol’s comparison against BFT, zkML, and optimistic approaches; the random small-quorum
Sampling Consensusdesign; deterministic execution across identical hardware/software; timeout/slashing logic; and the role of TEEs in protecting data and model weights during execution.
Node-side implementation surface
- https://github.com/AtomaAI/atoma-node
- Best source for what an actual Atoma execution node is expected to run.
- Used to confirm the node-operator flow: Sui client setup, registration against the Atoma smart contract, node badges and small IDs, public inference URLs, metrics endpoints, and deployment around OpenAI-compatible model-serving backends.
- Also useful for the plain statement that the Sui contract layer orchestrates payments, request authentication, load balancing, and other network coordination.
Proxy / access-layer implementation surface
- https://github.com/AtomaAI/atoma-proxy
- Best source for Atoma’s routing and API-access control plane.
- Used to confirm that Atoma exposes a distinct proxy layer for load balancing, request routing, authentication, local or cloud deployment modes, and hosted API-key-based access through
cloud.atoma.network/api.atoma.network. - Important because it reveals that user access and routing are not only properties of the node mesh; they are also mediated by a separate proxy service.
Onchain coordination layer
- https://github.com/AtomaAI/atoma-contracts
- Used to confirm the existence of a dedicated smart-contract repository for Atoma’s Sui-based coordination layer.
- The repo README is minimal, but together with the node/proxy docs it grounds the claim that node registration and broader network orchestration are contract-mediated rather than purely offchain.
Whitepaper / research materials
- https://github.com/AtomaAI/atoma-docs/blob/main/papers/atoma_whitepaper.pdf
- Official whitepaper location referenced directly by the node repository.
- Used to confirm that Atoma treats
Sampling Consensusas a canonical protocol primitive rather than only a docs-site description.
- https://arxiv.org/abs/2410.13752
- Companion research paper:
Privacy-Preserving Decentralized AI with Confidential Computing. - Used to confirm the formal privacy framing: TEEs as a practical alternative to current zkML limits for decentralized AI, especially when model weights and user data must stay confidential in a permissionless environment.
- Companion research paper:
Main takeaways from this pass
- Atoma is best understood as a sampled-verification plus confidential-computing stack, not as a generic decentralized GPU marketplace.
- The most useful analytical split is between execution nodes, proxy-side request routing and API access, Sui-mediated node admission / collateral / coordination, sampled verification, and TEE-backed private execution.
- Atoma is especially valuable in the corpus because its public materials reveal both a protocol architecture and a commercialization drift: decentralized-network docs and repos still exist, but the main website now foregrounds an enterprise confidential-AI platform. That makes the boundary between protocol, middleware, and operator-run service layers unusually legible.