Agent Architecture
1. What the architecture does
Prodia Agent architecture is responsible for turning a user’s industrial question into a controlled execution path and an actionable response.
2. Why one entry point does not mean one monolithic Agent
From the user’s perspective, Prodia can appear as one conversational Agent entry point. Internally, however, it is better understood as an orchestrated system that routes work across scenario logic, governed capabilities, context handling, memory, and knowledge.
That means the same conversational entry point may organize different kinds of capability according to the question, such as:
- production and business analysis
- quality and anomaly diagnosis
- fault and equipment investigation
- knowledge retrieval and experience enhancement
- result organization and explanation
The value of this design is that the system does not ask one monolithic Agent to do everything equally well. It lets the right capability participate in the right scenario.
3. Logical layers
| Layer | Responsibility | User value |
|---|---|---|
| Interaction layer | receive natural-language questions and support follow-up turns | users can keep analyzing through conversation |
| Routing and orchestration layer | identify intent, scope, and execution path | complex questions become structured steps |
| Capability module layer | apply domain methods and usage rules for different roles and scenarios | responses stay closer to operational reality |
| Scenario capability layer | organize output, quality, takt, fault, SPC, and other domain capability | analysis stays grounded in industrial language |
| Memory layer | preserve continuity, preference, and intermediate task state | fewer repeated clarifications across turns |
| Tool execution layer | invoke governed business tools and structured capabilities | answers stay grounded in real execution |
| Knowledge enhancement layer | bring in cases, rules, glossary, and organizational experience | improves explanation and reuse |
| Governance layer | keep permissions, safety, and trust boundaries intact | the system stays reliable and controllable |
4. Why the architecture is now better explained as three collaborative layers
If we look deeper into runtime behavior, the current Prodia architecture is no longer best explained as “prompt-first”. It is better understood through three layers:
| Layer | Focus | What it does |
|---|---|---|
| Harness Engineering | runtime, tools, sessions, events, state, and governance | defines the execution world in which Agent operates |
| Context Engineering | history, memory, knowledge, time, tool boundary, and budget | decides what this turn is allowed to see |
| Prompt Engineering | role, rules, mode switch, skill visibility, and output contract | decides how the turn should behave and respond |
In simple terms:
- Harness builds the execution world
- Context selects what the turn can see
- Prompt shapes how the turn behaves and explains
For a deeper explanation, continue with Three-layer Architecture, Context Engineering, and Prompt Engineering.
5. Relationship overview
Seen through the three-layer lens:
- Harness guarantees that sessions, tools, state, and execution boundaries really exist
- Context decides how session context, memory, and knowledge enter the turn
- Prompt turns those inputs into stable behavior and output constraints
6. Architectural principle
The Agent should not behave like a free-form generator.
Its job is to organize reliable industrial execution under context and boundary, then turn the result into something operations teams can use immediately.