Agentic search has moved from fixed RAG pipelines into flexible agent harnesses with access to a set of search tools: keyword search (bm25, grep regex) and semantic search.

When you upload a collection of unstructured documents to LlamaParse, we expose all these tools for agents to access.

Come check out our webinar on June 30th where we explore all these different tools and identify which ones work the best for agentic search:

https://landing.llamaindex.ai/retrieval-harness…

Building Agentic Retrieval with LlamaParse Index

Source: https://landing.llamaindex.ai/retrieval-harness June 30th | 9 AM PST | Register to attend

You have the right model, but do you have the right context? For most teams the answer is no — and the reason is in knowledge harness and retrieval. Teams are split: some threw out their vector databases entirely, grep and file skills only. Others bet everything on embeddings. So which side is right?

Turns out — both are. Semantic search gives agents a fast first pass over large corpora. Grep and file reads give them the precision to verify, dig deeper, and recover when the top-k chunks cut off mid-answer.But stitching grep and semantic search into a single harness is harder than it looks at scale: server-side search across multi-tenant document corpora, index freshness, permission boundaries, and complex file formats that agents can actually navigate (with text, layout, metadata, and page screenshots) rather than hallucinate through.

We built this harness into LlamaParse Index: semantic search, server-side grep, and file-level navigation in one reasoning loop. Join LlamaIndex Head of Engineering George He for a look under the hood—the architecture decisions, the dead ends, and a live demo of an agent reasoning across multiple indexes on a real enterprise task.

What we’ll cover:

• Why the grep-vs-embeddings debate is a false binary: what benchmarking both approaches taught us, and why the answer changes with corpus size
• The harness, end to end: grep, directory listings, and direct file reads as first-class agent tools, composed with hybrid search and reranking; plus multimodal file objects that give agents visual context for tables where text extraction fails
• Live demo: an agent navigating multiple indexes — search, grep, read, verify, answer

LlamaIndex 🦙 (@llama_index): Vector databases or pure grep? Teams are split on the right retrieval architecture for agents.
⁣ ⁣ The reality? You need both. Semantic search for a fast first pass; grep and file reads for surgical precision when top-k chunks cut off mid-answer.
⁣ ⁣ On June 29, our Head of