WarpGrep as alternative to RAG for code search

[bhaktatejas922]

Embedding-based retrieval has a mathematical ceiling for code queries (DeepMind, arxiv 2508.21038): recall@100 drops below 20% at scale, and embeddings go stale in active repos.

WarpGrep is an agentic alternative. It runs grep/read/list in an isolated context window with 8 parallel calls per turn (up to 4 turns), returning file:line-range spans. No indexing, no embeddings, no staleness.

SWE-bench Pro with Opus: 39% fewer input tokens, 26% fewer turns, 10% higher solve rate.

OpenAI-compatible API. Could complement or replace the RAG pipeline for code search.

Docs: https://docs.morphllm.com/sdk/components/warp-grep/direct
Benchmarks: https://morphllm.com/benchmarks