Knowledge Base RAG

Knowledge Base / RAG Capability

When to Use

Use this capability for indexing documents, retrieving relevant context, building grounded answers, maintaining a project knowledge base, and citing evidence from approved files.

Preferred Implementation

• Use local indexing, embeddings, vector stores, keyword search, or connector search depending on available tools.
• Use project-knowledge-graph when the task depends on relationships, dependency paths, graph traversal, change impact, or multi-agent continuation rather than semantic similarity alone.
• Prefer source-grounded retrieval before generation.
• Use Computer Use only when the knowledge source is accessible only through a visual UI and user approves.

Inputs

• Approved document set.
• Query or task.
• Retrieval scope.
• Citation requirements.
• Sensitive data rules.

Outputs

• Retrieved Evidence.
• Grounded Answer.
• Source Citations.
• Knowledge Gap Notes.
• Index or Refresh Report.

Security and Ethics

• Do not index secrets, credentials, raw customer-sensitive data, or private files without approval.
• Do not present unsupported statements as sourced facts.
• Keep citations tied to actual retrieved evidence.
• Use project-memory-learning for approved preferences, decisions, and experiment history; use this capability for document retrieval and grounded evidence.
• Keep each project's index and graph separate by default. Do not use a global index to silently connect private projects.