iterative-retrieval

Iterative retrieval solves unpredictable subagent context needs by progressively dispatching, evaluating, and refining search queries to dynamically narrow down relevant codebase files without requiring upfront knowledge.

Overwhelming subagents with entire codebases causes "context too large" errors. Iterative retrieval optimizes token usage by iteratively refining search queries to extract only the most relevant code snippets, preventing context overflow in multi-agent workflows.

You need a RAG-like retrieval pipeline when debugging complex code flows, such as authentication, where relevant files cannot be predicted upfront. Iterative retrieval progressively discovers and retrieves specific files without overwhelming the agent with the entire codebase.

Build a RAG pipeline using iterative dispatch, evaluation, and refinement loops that dynamically adapt search patterns and keywords based on initial evaluations to progressively narrow down relevant codebase context.

Yes, iterative retrieval works for debugging complex authentication flows by intelligently discovering and retrieving only relevant files like auth.ts, session-manager.ts, and jwt-utils.ts. It progressively refines context without loading the entire codebase.

Iterative context refinement requires multiple dispatch and evaluation loops, which may increase latency before retrieving the final codebase context. It is designed for complex, unpredictable context needs rather than simple, direct file lookups in multi-agent workflows.