What is Conditional Search Tool?
Conditional Search Tool is the Scavio binding pattern where the tool schema includes explicit needs_fresh_data criteria the model evaluates before invoking. Instead of the tool firing on every turn, the model checks whether the answer depends on information that could have changed, whether the context window already contains it, and whether a cheaper tool could satisfy. Teams that adopt the pattern typically cut Scavio call volume 40 to 70% without degrading answer quality, which lowers both spend and latency.
Example Response
JSON
{ "called": true, "reason": "query references 2026 pricing, context lacks freshness", "results": [{ "title": "...", "url": "..." }] }Use Cases
- High-volume research agents that cannot afford per-turn search
- Chat products optimizing for latency
- Multi-step agents with long context windows
- Cost-sensitive production deployments
Why Conditional Search Tool Matters
Always-on search binds waste tokens and credits on turns where the model already knows the answer; conditional binding matches spend to actual need.
LangChain Example
Drop conditional search tool data into your LangChain agent in a few lines:
Python
tool = ScavioSearch(invoke_when="needs_fresh_data").bind(llm)