# Traditional Approach
from openai import OpenAI
from anthropic import Anthropic
client\_openai = OpenAI(api\_key="...")
client\_anthropic = Anthropic(api\_key="...")
# Manually choose which model to use
response = client\_openai.chat.completions.create(...)
# manadia Approach
from manadia import ManadiaClient
client = ManadiaClient(api\_key="...")
response = client.chat.completions.create(
messages=\[...],
quality\_requirement="high", # Automatically selects the optimal model
# or
cost\_limit=0.001, # Selects the lowest-cost model that meets the requirements
# or
latency\_slo=500, # Selects the fastest available model
)
| | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | The value of AI API Routing lies in significantly reducing the complexity of integrating multiple AI models while continuously accumulating model invocation data and AI workload profiles for the manadia network. #### 6.2 Agent Trace Agent Trace is the core observability infrastructure of the manadia AI-Native Compute Coordination Network. Built upon distributed tracing architecture, Trusted Execution Environments (TEE), and on-chain proof technology, Agent Trace serves as the foundation for making AI Agent execution observable, reproducible, and auditable. It is the core infrastructure supporting the manadia Trustworthy AI framework, ensuring that execution results can be reproduced and independently verified. **Product Positioning** Agent Trace is an observability and auditing platform covering the entire lifecycle of AI Agents—from task requests, scheduling, execution, and settlement. It serves both enterprise AI applications for operational monitoring and debugging, as well as the manadia Trustworthy AI Prediction Model and quantitative trading agents for execution replay, risk tracing, and compliance auditing. **Core Technical Architecture** Agent Trace adopts a six-layer distributed tracing architecture to ensure data integrity and immutability. * Collection Layer — Non-intrusive instrumentation integrated with AI Workloads, Agent Runtime, and MCP modules. * Standardization Layer — Generates globally unique Trace IDs associated with model hashes, data fingerprints, and other trusted identifiers. * Storage Layer — Time-series databases combined with distributed storage, with critical execution hashes anchored on-chain. * Analytics Layer — Computes latency, cost, success rate, and other metrics in real time while supporting anomaly detection and bottleneck analysis. * Visualization Layer — Displays execution traces through the AI User interface and the Trustworthy-AI Auditor. * Audit Interface Layer — Provides standardized interfaces for blockchain auditors and institutional compliance systems. **End-to-End Trace Data Collection** Agent Trace captures every critical event throughout the entire execution lifecycle of an AI Agent. Every record includes timestamps, TEE execution proofs, and data fingerprints. Collected events include: * User requests, model invocations, and Prompt versions * Tool Calls and MCP invocations * Data source access, intermediate outputs, and execution actions * Errors, retries, Token consumption, and final outputs **Core Technical Capabilities** * Distributed End-to-End Sampling — Adaptive sampling strategies with full sampling for critical tasks while maintaining system performance. * End-to-End Latency Analysis — Breaks down execution latency across the entire call chain and visualizes performance bottlenecks. * Accurate Root Cause Analysis — Automatically categorizes failures and distinguishes between model, tool, data, and compute node issues. * Multi-Dimensional Cost Attribution — Calculates Token and compute costs by Agent, model, user, and task to support settlement. * Trusted Execution Proof — Anchors critical execution hashes on-chain and combines them with TEE proofs to ensure tamper-proof, verifiable execution. * Real-Time Risk Monitoring — Integrates with the risk engine to detect abnormal execution and trigger circuit breakers or degradation mechanisms. **Application Scenarios** * Enterprise AI Applications — Agent debugging, performance optimization, and compliance auditing. * Trustworthy AI Prediction Models — Quantitative strategy replay, execution verification, and risk tracing. * Developer Ecosystem — AI application debugging, API troubleshooting, and cost management. * Institutional Compliance — AI execution auditing and immutable data retention to satisfy global regulatory requirements. #### 6.3 Vertigas Oracle Vertigas Oracle is the decentralized trusted data gateway of the manadia ecosystem. Built on a Dual-Lane Architecture, it combines AI-driven automated data reporting with human challenge and arbitration mechanisms to deliver highly trusted, low-latency, and arbitrable external data and event resolution services. It serves as the trusted bridge connecting on-chain execution with real-world information. **Product Positioning** As the dedicated decentralized Oracle and data network of the manadia ecosystem, Vertigas supports both high-frequency market data and complex real-world event resolution. It provides standardized data inputs and settlement verification for the Trustworthy AI Prediction Model, Prediction Markets, and AI Agents, while maintaining data integrity through economic incentives and reputation mechanisms. **Core Technical Architecture** Vertigas adopts an independent dual-lane transmission architecture with a unified on-chain gateway for data aggregation and settlement. * Fast Lane — Designed for high-frequency price feeds, supporting sub-millisecond updates, multi-signature aggregation, circuit breaker protection, and replay attack prevention. * Slow Lane — Designed for complex event resolution, supporting AI-generated proposals, optimistic challenge windows, staking guarantees, and dispute arbitration. * Unified Dual Gateway — A Price Feed Gateway and an Event Oracle Gateway jointly handle price validation and final event settlement. **Core Capabilities** * Full Lifecycle Data Source Management — Registration, classification, digital signatures, timestamp verification, and reputation scoring. * AI-First Resolution — AI Oracle nodes provide default event outcomes, enabling second-level response times. * Human Challenge Mechanism — A 48-hour challenge window allows incorrect results to be disputed and corrected, with successful challengers receiving rewards. * Falsification and Slashing — Invalid or incorrect data triggers staking penalties while simultaneously updating data source reputation scores. * Node Economics — Validator nodes stake tokens to participate in verification services and share protocol fee revenue. **Ecosystem Integration** * Provides trusted high-frequency market data, on-chain information, and event signals for the Trustworthy AI Prediction Model. * Serves as the foundational layer for result validation, on-chain settlement, and dispute arbitration within the manadia Prediction Market. * Delivers compliant, verifiable, and tamper-proof external data inputs for AI Agents across the entire manadia ecosystem. #### 6.4 manadia Prediction Market The manadia Prediction Market is the primary application scenario for trustworthy predictive intelligence. It adopts on-chain clearing and automated settlement, supporting conditional events, binary predictions, and range-based predictions. Together with the AI Prediction Model, it forms a positive feedback loop of Data → Prediction → Verification → Iteration. The platform supports event forecasting, market direction prediction, on-chain event prediction, and macroeconomic trend forecasting. The AI Prediction Model and the Prediction Market maintain a bidirectional relationship. The AI model provides probability analysis, risk assessment, and event interpretation for the Prediction Market, while the Prediction Market supplies collective expectations, market pricing signals, and event outcome feedback to continuously improve the AI model. This relationship enables manadia to build not merely a prediction market, but an integrated prediction system powered by AI models, data sources, Oracles, and user participation. **Prediction Event Lifecycle Management** The Prediction Market adopts a standardized on-chain event creation and management mechanism, providing complete traceability from event creation to settlement. Event Creation and Admission Authorized ecosystem participants may create prediction events by submitting core parameters on-chain, including the event title, description, settlement rules, prediction deadline, and result evaluation time. All information is permanently recorded on-chain and cannot be modified. The system automatically validates scheduling logic (ensuring the evaluation time is later than the prediction deadline). Once validated, the event becomes publicly available for participation. Event Types The platform supports three standardized prediction formats: * Binary outcome predictions * Range-based predictions * Conditional event predictions These event types are applicable to cryptocurrency market movements, on-chain events, Real-World Assets (RWAs), macroeconomic indicators, and many other prediction scenarios. All event rules and settlement criteria remain fully transparent. Liquidity Formation Market pricing and liquidity emerge organically through user participation, without centralized liquidity pools. Information pricing and market equilibrium are achieved through on-chain interactions. **Core Interaction Features** The Prediction Market provides lightweight, blockchain-native user interactions, with every operation executed through smart contracts. * Prediction Participation — Users connect their wallets, select an event and prediction direction, and participate by submitting on-chain transactions without requiring authorization from any centralized platform. * Dynamic Position Management — Before the prediction deadline, users may adjust or close their positions, with every modification executed on-chain according to real-time market conditions. * Traceable Interaction History — The system permanently records participation history, position changes, and event status updates, allowing verification at any time through on-chain transaction hashes. **Event Resolution and On-Chain Settlement** Event resolution and settlement are entirely driven by the Veritas Oracle Dual-Lane Architecture, enabling a trust-minimized, dispute-resistant, and fully automated settlement process. Result Determination After an event closes, the Fast Lane processes high-frequency price events while the Slow Lane resolves complex real-world events. By default, AI Oracle nodes produce the initial result. A 48-hour on-chain challenge period allows participants to dispute the outcome. Following dispute resolution, the final settlement result is confirmed. Automated Settlement Once the final result is confirmed, smart contracts immediately execute automated settlement, distributing assets and clearing positions on-chain without manual intervention or operational delay. Settlement records are permanently stored on-chain. **Core Technical Features** * Verifiable Execution — Every stage, including event creation, interaction, resolution, and settlement, is recorded on-chain, making the entire lifecycle verifiable and traceable. * Dispute-Resistant Resolution — Combines AI-based judgment with human challenge mechanisms through the Veritas Oracle to prevent manipulation and disputes. * Trust-Minimized Architecture — Eliminates centralized control by relying on smart contracts and Oracle infrastructure to ensure fairness. * Compliance-Oriented Extensibility — Integrates with the AurumX compliance framework to support institutional participation and cross-market prediction assets. * AI Closed-Loop Optimization — Maintains bidirectional data exchange with the manadia Trustworthy AI Prediction Model. AI provides prediction signals, while the market supplies verified event outcomes that continuously improve model inference. #### 6.5 Potion Potion is the trusted behavior certification and rights credential system designed for long-term participants within the manadia ecosystem. Built on Zero-Knowledge Proof (zk-SNARK) technology, its core value lies in transforming users' comprehensive participation activities into structured, on-chain, and verifiable records. These immutable participation records provide the trusted foundation for ecosystem identity, permission management, and service personalization. **Product Positioning** As the standardized participation infrastructure of the manadia ecosystem, Potion focuses on trusted behavior recording, cross-scenario verification, and long-term participation management. It connects compute contribution, model usage, data contribution, and ecosystem interactions into a unified trusted identity layer. **Core Technical Features** * Participation eligibility verification based on zk-SNARK, with verification proofs of approximately 200 bytes, balancing privacy protection and verification efficiency. * Supports long-term participation modeling over 360+ days, enabling dynamic behavior tracking and natural reputation decay. * Cross-platform data integration, allowing participation records to be aggregated across both internal and external ecosystems. * End-to-end on-chain proof storage, with behavior record hashes anchored on-chain to ensure immutability and continuous verifiability. **Core Product Features** * Unified Ecosystem Identity Management — Binds wallet addresses to generate a unique ecosystem identity. * Multi-Dimensional Behavior Records — Structurally records compute contributions, model participation, data provision, and ecosystem activities. * Rights Status Visualization — Displays participation dimensions, contribution levels, permission status, and user profiles in real time. * Long-Term Reputation Modeling — Builds an ecosystem reputation system based on historical participation to support service matching and access control. * Lightweight Trust Verification — Uses Zero-Knowledge Proofs to verify eligibility without exposing users' private information. **Technical Value** Potion transforms unstructured user participation into standardized, verifiable, and extensible on-chain records, establishing a unified trusted identity and behavior layer for the ecosystem. It provides standardized support for compute scheduling priorities, model service permissions, and data access authorization, further strengthening the identity infrastructure of the manadia Trustworthy AI ecosystem. #### 6.6 manadia Pay manadia Pay is an AI-native payment and settlement platform built upon the global compliance framework of AurumX. Designed for AI compute consumption, model subscriptions, data services, API usage, and other AI-native scenarios, it provides a unified compliance solution combining on-chain payments, fiat connectivity, and enterprise-grade settlement. It serves as the primary payment gateway for value circulation across the manadia ecosystem. **Product Positioning** As the native payment and settlement layer of the manadia network, manadia Pay addresses the challenges of large-scale AI service payments by integrating on-chain stablecoin settlement with global fiat payment channels. It provides secure, compliant, and efficient payment and reconciliation services for individuals, developers, and enterprises, completing the value circulation loop of the AI compute economy. **Core Technical Features** * Global Compliance Framework — Built upon the AurumX compliance infrastructure, leveraging Hong Kong SFC licenses and U.S. FinCEN MSB registration to support multi-jurisdictional operations. * Dual Payment Channels — Combines real-time stablecoin settlement with fiat on-ramp and off-ramp services to accommodate both retail and institutional users. * Enterprise-Grade Settlement — Supports bulk payouts, intelligent revenue sharing, and automated reconciliation for large-scale business operations. * Open API Integration — Provides standardized payment and reconciliation APIs that integrate seamlessly with AI API Routing, data subscriptions, and compute billing services. **Core Product Capabilities** * AI-Specific Payment Services — Supports settlement for compute rental, model subscriptions, data services, and API consumption. * Multi-Currency Payment Support — Compatible with multiple stablecoins while supporting 30+ major fiat currencies for global users. * Enterprise Billing Management — Provides monthly billing, usage analytics, and cost reporting for enterprise customers. * Developer Revenue Distribution — Enables automated revenue sharing and payments for data providers, model developers, and compute node operators. * End-to-End Reconciliation — Anchors payment, settlement, and revenue-sharing records on-chain while providing standardized reconciliation documents and transaction verification. manadia Pay establishes a compliant value transfer infrastructure for the AI compute economy, transforming compute consumption, model usage, and data service fees into standardized, automated, and verifiable payment and settlement processes. It addresses fragmentation, compliance challenges, and settlement inefficiencies within the AI ecosystem while supporting the large-scale commercialization of the manadia ecosystem. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://mana.gitbook.io/manadia/6.-product-ecosystem.md?ask=