Summary: Lilypad is best cataloged not as a generic GPU marketplace, but as a coordination protocol that tries to separate compute execution, verification, registries, and payment into distinct but interoperable layers. Its official litepaper and docs describe a system where job creators fund AI workloads, resource providers run them, solvers match jobs to compute, mediators optionally verify correctness, and onchain contracts handle staking, slashing, and settlement. The reusable mechanism insight is that Lilypad treats verifiability as a programmable layer rather than as a prerequisite that must fully constrain execution, which makes it a useful comparison class for decentralized compute, agent infrastructure, and crypto-native AI middleware.
What it does:
Lets users run containerized AI workloads across a decentralized GPU network through a CLI, inference API, web studio, and developer-facing module system
Lets resource providers contribute GPU or CPU capacity and earn rewards for completed jobs
Supports reusable job modules and a model/workload marketplace for publishing and monetizing AI pipelines
Describes a multi-role network with job creators, resource providers, module providers, solvers, mediators, and smart contracts coordinating job routing and payment
Exposes a broader stack named in the litepaper as Halo for identity/auth, Ophir for contract coordination, Ampli for registry/IP metadata, Anura for inference/API integration, and Atlas for the underlying compute layer
Key claims:
The litepaper defines Lilypad as a coordination protocol for intelligence infrastructure, emphasizing permissionless access to verifiable compute, model monetization, and cryptographically auditable execution for AI workloads
The docs homepage presents Lilypad as a full-stack modular AI services platform spanning a model marketplace, MLOps tooling, and a distributed on-demand GPU network
The litepaper explicitly says Lilypad decouples verification from execution so different job types can use different validation strategies rather than forcing one deterministic trust model on every workload
The docs query interface summarizes the network roles as job creators, resource providers, module providers, solvers, mediators, and smart contracts, which makes matching and audit policy part of the protocol’s real control surface
The GitHub org and main lilypad repository show substantial first-party code and module infrastructure rather than a thin token wrapper around outsourced compute
The docs position Lilypad as a demand engine for decentralized compute and storage plus a distribution layer for agentic frameworks and training networks, which makes it analytically adjacent to crypto-AI coordination systems rather than only DePIN supply marketplaces
The most important open question is not whether Lilypad uses decentralized hardware, but where practical authority settles across solver logic, registry curation, mediation/verification rules, and the contract layer that decides staking, slashing, and payments
Whitepaper: The main current technical source is the official litepaper/docs stack at https://docs.lilypad.tech/lilypad/research-and-vision/lilypad-litepaper; no standalone local PDF was pulled in this pass. The strongest operational snapshot is ../whitepapers/lilypad-primary-sources-2026-05-08.md.