Acronym Definition
CIG = Context Inferenced Generation
A generation paradigm where curated context vaults are accessed by inference engines to retrieve relevant knowledge, which then drives multimodal content generation.
CIG is the pattern of using inference to intelligently select context from structured knowledge stores (context vaults), then generating new content based on that retrieved context. The key differentiator: inference drives context selection, not just vector similarity or static caching.
A context vault is a curated, structured knowledge store containing:
- Agent memory (preferences, decisions, history)
- Skills and methods (how-to knowledge)
- Documentation and references
- Research and notes
| Pattern | Context Source | Selection Mechanism | Output |
|---|---|---|---|
| RAG | Document corpus | Vector similarity | Text |
| CAG | Pre-loaded cache | Static | Text |
| CIG | Context vault | Inference engine | Multimodal |
Context Vault → Inference Engine → Generator → Output
The inference engine semantically retrieves relevant context from the vault, then generators produce content in any modality.
- Curated context - Structured knowledge, not raw documents
- Inference-driven - Intelligent selection, not just similarity matching
- Multimodal - Generates text, images, audio, scripts, or other content
- Agent-centric - Designed for autonomous agent workflows
When an agent receives a request:
- Inference engine searches the context vault
- Relevant memory, skills, and docs are retrieved
- Generator produces response using that context
This is CIG: inference determines what context matters, generation creates from it.
Proposed term for the context-augmented generation taxonomy.