¶ Prompt Context Builder
The Prompt Context Builder orchestrates how knowledge is dynamically assembled and injected into AI prompts — ensuring relevance, security, and explainability.
¶ Overview
Fuse uses a composable system called the Prompt Composer to construct AI prompts using knowledge retrieved from:
- Embeddings
- Knowledge Graph traversal
- Tagged or scored Resources
- Workflow outputs
- Assistant or session scope
This allows context-aware, access-governed inference — improving model responses while maintaining traceability.
¶ Why It Matters
- Ensures only trusted, permitted context is used
- Enables dynamic enrichment of prompts during conversations or tasks
- Supports hybrid strategies: semantic + symbolic + source-based injection
- Enables explanation of where facts or context came from (provenance)
¶ Key Context Sources
| Source | Description |
|---|---|
 Embeddings |
Vectors retrieved by semantic similarity to the user query |
 Graph Nodes |
Relevant entities or concepts linked via traversal rules |
 Resources |
Filtered documents or live data exposed by Resource methods |
 Pipelines |
Runtime outputs injected from workflow or task execution |
Each source is filtered by scope, session, and policy.
¶ Composition Logic
- The system identifies the target assistant or scenario
- It selects scoped KnowledgeBases and trusted Resources
- Retrieves embeddings, traverses graphs, or queries live providers
- Applies filters: access control, recency, trust score, tags
- Assembles a composite prompt with structured content blocks
- Returns an explainable prompt with source metadata
¶ Use Cases
- AI Assistant reasoning with just-in-time context
- Preloading context into workflow-generated decisions
- Generating policy-aware answers with source references
- Augmenting assistant behavior based on task stage or history
¶ Features
| Feature | Description |
|---|---|
 Access Filtering |
Removes any knowledge the user/agent is not permitted to use |
 Explainability |
Tracks which resources or facts powered the AI response |
 Dynamic Re-evaluation |
Rebuilds prompt context on new input, task, or role |
 Fallback & Defaults |
Allows predefined static context if dynamic sources fail |
¶ Future Enhancements
- Assistant-specific prompt composition policies
- Weighted context ranking by trust, source type, or usage history
- Support for LLM feedback-driven prompt refinement
- Inference-time graph traversal hints and caching
Next: Traversal & Query API — Discover how to query and navigate knowledge graphs for context injection and reasoning workflows.