> For the complete documentation index, see [llms.txt](https://mana.gitbook.io/manadia/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mana.gitbook.io/manadia/2.-design-principles.md).

# 2. Design Principles

The technical architecture of manadia is built upon the following design principles.

#### 2.1 AI Workload First

The primary object of scheduling in manadia is not raw GPU resources, but AI workloads. An AI workload may consist of model inference, context processing, data retrieval, tool invocation, strategy execution, replay records, and audit logs. The system must understand the nature of the task before determining which compute resources, models, data sources, and execution paths should be used.

| **Dimension**          | **Traditional Approach**                                  | **Workload-First Approach**                                                                     |
| ---------------------- | --------------------------------------------------------- | ----------------------------------------------------------------------------------------------- |
| Scheduling Information | CPU utilization: 80%, GPU utilization: 60%, Memory: 32 GB | Requires GPT-4V + 10-second latency limit + low cost                                            |
| Routing Decision       | Automatically selects the least-loaded node               | Selects the optimal combination of model and execution environment based on capability matching |
| Failure Handling       | Node failure → Request fails                              | Node failure → Automatically switches to backup models and regions                              |
| Cost Control           | No cost awareness                                         | Knows how much the task costs and why                                                           |

#### 2.2 Model-Agnostic

manadia is not tied to any single model provider. The network can integrate model APIs from OpenAI, Anthropic, Gemini, DeepSeek, Grok, and other providers, while also supporting proprietary models, enterprise private models, locally deployed models, and future domain-specific models. The model routing system dynamically selects the most appropriate model based on task requirements, cost, latency, context length, stability, and trust level.

**Routing Strategy — Dynamically Selected Based on Task Requirements:**

* High accuracy required → GPT-4o
* Low cost with sufficient accuracy → Claude 3.5 Haiku
* Local privacy required → Local open-source model
* Low latency required → Nearest inference node

#### 2.3 Trust by Execution, Not by Claim

Trustworthy AI should not rely solely on claims made by the project. manadia establishes trust through the execution process itself: model versions are recorded, data sources are traceable, execution environments are isolated, strategy parameters are auditable, and results are reproducible. Trustworthiness does not mean that a model is always correct; rather, it means that the process by which the model produces its results can be verified.

The trustworthiness of manadia is built upon execution observability:

* Recordable Model Versions — Every inference records the model version, parameters, temperature, top\_k, and other execution settings.
* Traceable Data Sources — Records exactly which data source and snapshot were used as model input.
* Replayable Tool Calls — Records which tools were invoked, the input parameters, and the returned outputs.
* Isolated Execution Environment — Sensitive tasks are executed within a Trusted Execution Environment (TEE).
* Auditable Results — The complete decision-making workflow can be verified by external auditing systems.

#### 2.4 Modular Architecture

manadia adopts a decoupled modular architecture. Compute integration, model routing, Agent Runtime, data sources, Oracle services, TEE execution, strategy engines, auditing systems, and product applications are all independently decoupled. This modular architecture enables continuous expansion with new models, data sources, strategies, and products.

* New Features Without Architectural Changes — Adding new models, data sources, or tools requires no modification to the underlying architecture.
* Fault Isolation — A failure in one node does not affect the rest of the system.
* Localized Performance Optimization — Individual layers can be optimized independently without impacting other components.

#### 2.5 Product-Driven Infrastructure

manadia does not build infrastructure for the sake of narrative. Instead, it validates the value of its infrastructure through real-world products. The Trustworthy AI Prediction Model, AI API Routing, Agent Trace, and Vertigas are all productized implementations of the underlying compute coordination network.

manadia follows a product-first methodology by starting from high-value application scenarios and working backward to derive infrastructure requirements.

For example:

Application Scenario:\
**AI-driven financial prediction and automated trading.**

**Requirements of This Scenario:**

* Low latency (<1 second)
* High prediction accuracy (to generate excess returns)
* Comprehensive auditing (verifiable on-chain)
* Multi-source data coordination (market data + on-chain data + prediction markets)
* Multi-tool integration (wallets, exchanges, risk management systems)

**Derived Infrastructure Requirements:**

* Real-time data pipelines
* Low-latency model routing
* Agent execution capabilities
* Trusted Execution Environment (TEE)
* End-to-end execution traceability


---

# 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/2.-design-principles.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.