¶ Runtime Usage & Injection
The Runtime Usage & Injection capability enables Fuse to deliver precise, governed knowledge into real-time execution flows, prompts, decisions, and AI interactions.
¶ Overview
At runtime, Fuse intelligently injects knowledge into:
- Workflow task steps
- AI Assistant prompts
- Policy evaluation or validation rules
- Dynamic UI elements
- External system integrations
This ensures decisions and inferences are always context-aware, scoped, and explainable.
¶ What Can Be Injected
- Extracted paragraphs or summaries from KnowledgePages
- Related KnowledgeNodes and linked concepts
- Embedding-matched results with contextual relevance
- Dynamic outputs from Workflow-based Resources
- Live responses from Knowledge Providers
All injected content is filtered by role, session, and trust level.
¶ Injection Targets
| Target | Description |
|---|---|
 AI Prompts |
Adds scoped, relevant knowledge as part of RAG orchestration |
 Workflow Tasks |
Inserts content into forms, instructions, or validation rules |
 Looped Pipelines |
Feeds retrieved knowledge into iterative executions |
 Human Decisions |
Augments approvals or assignments with inline knowledge |
 External Consumers |
Enables Fuse-powered APIs to serve external assistants or tools |
¶ Injection Behavior
- Governed by
PromptComposer, workflow bindings, or task rules - Access-controlled by Resource and session context
- Configurable per KnowledgeBase, domain, or assistant
- Version-aware, trust-scored, and explainable
- Caches recent resolutions when enabled by policy
¶ Scenarios
- Injecting live policy guidance into an approval step
- Populating an assistant prompt with domain-specific context
- Providing users with a rationale graph during review
- Auto-resolving related procedures based on task inputs
- Injecting citation-traceable facts into chatbot responses
¶ Governance at Runtime
Fuse ensures:
- No unauthorized knowledge is exposed
- All injections can be traced back to source Resource(s)
- Context is dynamically refreshed as input evolves
- Knowledge lifecycle (version, trust, validity) is respected
¶ Future Capabilities
- Live knowledge overlays in the UI
- Runtime injection policies per assistant/task/workflow
- Smart knowledge routing based on decision impact
- Injection testing and visibility tools for builders
Next: Developer Examples — Explore example implementations and usage patterns for building, linking, and using knowledge in real-time.