85 agentic tools audited for security incidents · Talk to the lab →
MotionticBook a call
Glossary

What is agentic search?

Agentic search is a retrieval approach in which an AI agent runs its own searches, greps and file reads over several steps, choosing each next query from what it has already found, instead of receiving pre-retrieved passages.

An agent works in a loop, so it can search, read what comes back, and search again with a sharper query. Agentic search uses that loop for retrieval. The agent holds lightweight references, such as file paths, saved queries and links, and loads the actual content only when the next step needs it.

Anthropic calls this just-in-time retrieval. It is how Claude Code finds its way around a repository, with glob and grep rather than a pre-built index.

The tools an agent searches with

  • File tools

    List, glob, grep and read. Cheap, always current, and strong on structured corpora such as code.

  • Index search

    A hybrid or vector index exposed as a tool, for large corpora where wording differs from the query.

  • Database queries

    SQL or API calls against systems of record, when the answer is a record rather than a passage.

  • Retrieval subagents

    A separate agent that searches in its own context and returns only what it found. Chroma's Context-1, released on 2026-03-26, is a 20B open-weight model trained to search over multiple turns and prune its own context.

What it costs and what it buys

Agentic search needs no index to build or keep in sync, reads the current version of every file, and can follow a lead across several hops. It spends more tool calls, tokens and time per question than a single retrieval.

Anthropic's Agent SDK guidance is to start with agentic search over the filesystem and add semantic search only if needed. Cursor's data points the other way for large codebases: adding semantic search raised answer accuracy by 12.5% on average. The full comparison is on the RAG vs agentic search page.

How Motiontic's engineers apply it

Motiontic's engineers give agents a small set of retrieval tools, each with a clear description of what it is good for, and let the agent choose per step. Tool outputs are paginated and trimmed so a broad grep cannot flood the context, and each tool stays only if evals show it raises task success.

Common questions

Is agentic search the same as RAG?

No. Classic RAG retrieves once before the model answers. Agentic search lets the agent retrieve repeatedly while it works, deciding what to look at next from what it has found.

What is just-in-time retrieval?

Anthropic's term for keeping lightweight references, such as file paths or queries, in context and loading the full data through tools only at the step that needs it.

Does agentic search need a vector database?

Not necessarily. Grep and file tools need no index. Many production agents still include an indexed search tool for large or conceptual corpora alongside the file tools.

Why is agentic search slower?

Each search, read and follow-up is a separate tool call and model step, so a question can take several round trips instead of one retrieval.

Running agents that have to hold up?

Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.