How the Agent Operates
How the Agent Operates (Perception → Reasoning → Action)
MindAI Agents are designed to behave autonomously by following a structured intelligence loop composed of three stages: Perception, Reasoning, and Action. This design allows agents to dynamically respond to inputs, make decisions using AI models, and execute verified outputs onchain using Web3 logic.
Based on the MindAI Orchestrator Architecture we derive that the Perception and Action are part of the External Inputs and Orchestration Core layers while the Reasoning is part of the Model & Reasoning layer.
In a simplified view, the orchestration of this behavior is achieved using Workflow Studio, a low-code environment that integrates with AI and Web2/Web3 action nodes.
1. Perception
Perception is the agent’s ability to receive signals from the world. These can come from user requests, external events, smart contract data, or off-chain services.
In Workflow Studio, this is handled using:
Trigger Nodes — monitor onchain and off chain events.
Webhook Nodes — capture HTTP requests, form submissions, or API calls.
Examples:
Detect a Twitter mention to activate a KOL agent.
Receive a new DAO proposal submission via webhook.
Monitor a wallet balance falling below a threshold.
2. Reasoning
Reasoning allows the agent to interpret context, apply logic, and decide on the next action. This layer is powered by Google Cloud AI, particularly:
Vertex AI LLMs (e.g. PaLM, Gemini) for language understanding and strategic decision-making.
Imagen for image generation (e.g. marketing content).
Speech-to-Text / Audio-to-Gesture pipelines for multimodal input analysis.
The agent interprets intent and routes decisions to the appropriate output logic.
Examples:
Classify if a user’s message should trigger a paid service.
Generate custom visual content for a marketing campaign.
Decide whether to accept or reject a smart contract proposal.
3. Action
Action is where agents interact with the external world—most often onchain. Using Action Nodes in Workflow Studio, agents can:
Execute smart contract functions
Transfer or stake tokens
Mint NFTs or generate invoices
Store files to IPFS or update databases
These actions are secure, auditable, and permissionless.
Examples:
Transfer $MINDAI to a user after a successful evaluation.
Mint a proof-of-contribution NFT.
Execute a buy order on a decentralized exchange.
Workflow Summary
[Trigger or Webhook]
↓
[Google Cloud Reasoning (LLM, Imagen, etc.)]
↓
[Action Node (e.g. mintNFT, transferToken, executeSmartContract)]
↓
[Onchain Result validation via MindAI Trust Protocol]Last updated