iterative-retrieval

Iterative retrieval solves context limits for multi-agent workflows by progressively narrowing relevant information from large codebases. It uses a 4-phase loop to score relevance and dynamically expand keywords, ensuring subagents get precise context without exceeding token budgets.

Iterative retrieval works by running a 4-phase loop: DISPATCH, EVALUATE, REFINE, and LOOP. It scores relevance, applies exclusions to filter noise, and extracts dynamic patterns to progressively narrow context for code search and information retrieval tasks.

Yes, iterative retrieval is designed for multi-agent workflows and analytics tasks. It optimizes agent context by feeding subagents just enough precise information, which is critical for data analysis tasks operating within limited token budgets.

The best way to prevent irrelevant results is using a retrieval loop with relevance scoring and dynamic pattern extraction. By applying exclusions during the evaluation phase, it guards against irrelevant results and ensures only highly relevant context is kept.

Yes, you need to provide an initial task description to trigger iterative retrieval for the target project. This description acts as the starting point for the loop to begin narrowing down relevant information from your codebase or documents.