Knowledge

Build persistent Starfire AI reference collections from files, repositories, indexed sources, and retrieval.

Knowledge

Knowledge is Starfire AI’s persistent reference layer. It is designed for information that should be reusable across conversations, projects, research, agents, and other workflows.

What a knowledge source can represent

Depending on the active feature set, Knowledge can be built from sources such as:

• uploaded documents
• project files
• repositories
• URLs or web content
• structured internal material
• other indexed sources supported by the platform

Ingestion lifecycle

A durable knowledge workflow usually has several stages:

Source added
    ↓
Text/content extraction
    ↓
Chunking + metadata
    ↓
Embeddings / indexing
    ↓
Retrieval-ready
    ↓
Query / reranking
    ↓
Relevant context returned

A file existing in storage does not necessarily mean it is immediately ready for semantic retrieval.

Retrieval instead of full injection

Knowledge bases can grow far beyond a model’s context window. Retrieval selects relevant pieces for a task rather than sending the entire collection to the model.

This is the foundation of Starfire’s RAG architecture.

Knowledge boundaries

Knowledge can be personal, project-owned, or organization-owned depending on how the source was created. Access should follow the owning context and associated permissions.

Source freshness

Persistent knowledge can become stale. A strong knowledge system needs source metadata, indexing status, and administrative tools for retrying failed or outdated ingestion.

Operational visibility

Control Center’s Knowledge administration is designed to expose operational metadata such as document counts, chunks, embeddings, index failures, stale sources, and storage without making private content an unrestricted admin browsing surface.

RAG & retrieval

Understand chunking, embeddings, retrieval, and reranking.

Repositories

Use code repositories as structured project knowledge.