Custom Agent Development
While Met.AI’s decentralized marketplace hosts a wide array of general-purpose AI agents, there are cases where users—especially enterprises, researchers, or domain-specific builders—cannot find an existing agent that fully meets their unique requirements. In such scenarios, the platform offers a robust pathway for Custom Agent Development, enabling users to commission tailor-made AI agents that align precisely with their objectives, data, and operational constraints.
This mechanism ensures that intelligence on Met.AI is not limited by what's already available, but can be collaboratively created to serve emerging and specialized demands. The process is transparent, structured, and entirely governed by smart contracts to ensure accountability and mutual protection for both the commissioner and the developer.
Custom agent development on Met.AI unfolds in three coordinated stages:
Developer Discovery & Selection
Users seeking custom agents begin by browsing a pool of verified developers on the platform. Each developer profile includes:This visibility ensures users can make informed choices and select developers best suited for the desired agent functionality.
Technical expertise：Tags and certifications such as “Transformer-based NLP,” “Reinforcement Learning,” “Web3 automation,” or “Vision + Edge AI.”
Portfolio references：Links to agents previously published or tasks completed via Met.AI.
Reputation score：An aggregated on-chain rating based on task history, responsiveness, and client satisfaction.
Specification & Contract Formation
After identifying a suitable developer, the user and builder collaboratively draft the agent specification. This includes:Once the scope is agreed, a milestone-based smart contract is deployed to govern payments, deliverables, and timelines. Each milestone corresponds to a development stage (e.g., prototype, beta, final deployment), and funds are escrowed until successful delivery.
Functional Requirements：Core logic, expected behaviors, input/output formats, and use case definitions.
Training Data & Sources：Whether training will be done with user-provided data, public datasets, or existing model fine-tuning.
Execution Context：Agent deployment environment (on-chain, off-chain node, TEE enclave), and compute requirements.
Ownership & Licensing：Definition of IP ownership, usage rights, and whether the agent will remain private or be listed on the marketplace.
Agent Deployment & Integration
Upon completion, the custom agent is deployed either privately to the commissioning user or publicly on the marketplace, depending on the agreement terms. The agent inherits all Met.AI native features:The final output is not just a finished model, but a live, self-contained agent ready for secure, modular execution across the decentralized network.
Service model：Can be offered as a private tool, or made rentable/subscribable for third parties.
Smart contract logic：Governed by programmable rules for interaction, pricing, access control, and logging.
Interoperability：Fully compatible with the Multi-Agent Context Protocol (MCP) and reputation system.
The final output is not just a finished model, but a live, self-contained agent ready for secure, modular execution across the decentralized network.
Last updated