Skip to main content

Core Concepts Overview

rag_control is built around several core concepts that work together to provide governance, security and observability for RAG systems.

rag_control's Solution​

rag_control provides a layered approach to governance:

User Request
    ↓
[Organization Validation]
    ↓
[Document Filtering]
    ↓
[Query Embedding]
    ↓
[Document Retrieval]
    ↓
[Policy Resolution]
    ↓
[LLM Generation]
    ↓
[Enforcement Checks]
    ↓
Response or Denial

Core Concepts​

1. Policies​

Policies define how the LLM should behave:

  • Generation parameters (temperature, output length)
  • Citation requirements
  • Knowledge restrictions
  • Enforcement rules

Example: "Strict policy requires citations and prevents external knowledge"

2. Governance​

Governance applies organization-level rules:

  • Organization-specific policy overrides
  • Role-based access control
  • Data classification rules
  • Conditional policy enforcement

Example: "Acme Corporation always uses strict policy for their queries"

3. Filters​

Filters control document retrieval:

  • Data classification (public, internal, confidential)
  • Metadata-based filtering
  • User context validation

Example: "Only retrieve documents marked for this organization"

4. Adapters​

Adapters integrate rag_control with your infrastructure:

  • LLM (language model)
  • Embeddings (query encoding)
  • Vector Store (document retrieval)

5. Observability​

rag_control provides comprehensive observability:

  • Audit Logging: Track every request and decision
  • Distributed Tracing: Follow execution flow
  • Metrics: Monitor performance and compliance

Execution Flow​

  1. Org Lookup: Validate user's organization
  2. Document Filtering: Apply org-level filters
  3. Query Embedding: Encode the query
  4. Document Retrieval: Fetch relevant documents
  5. Policy Resolution: Determine applicable policy
  6. LLM Generation: Generate response with policy constraints
  7. Enforcement: Validate response against policy
  8. Audit: Log request and decision

Learning Path​

  1. Start here: You're reading it!
  2. Dig deeper: Policies
  3. Understand governance: Governance
  4. Learn filtering: Filters
  5. Integrate adapters: Adapters

See Also​