Hello,
I am developing an internal company AI assistant. We’d like to use Brave Search API’s llm context endpoint to act as a WebSearch tool for the agents.
As with all agentic tools, the implementation would partially retain some of the results in the chat history/context of a given thread for a period of time so the model can see what it searched for and work with the results across more than a single turn of conversation.
The results are cleaned and fully deleted when:
A. Context compaction kicks in - the history is overwritten to contain only the query and a stub that the tool was called, none of the results persist anywhere.
B. The chat is deleted by the user - all tool results are wiped as well.
The results would never be used for fine tuning, training, re-use or similar uses.
The sole reason they would be retained is so that the model can reason about the results and draft answers based on the facts contained in them across more than a single turn.
It would be possible to discard at the end of each turn - however, I would like to avoid this to avoid frequent LLM context cache invalidations and to avoid potential output quality impacts caused by compacting the results away too frequently (the model can re-research them if need be, but I still prefer keeping that as a last resort when context pressure requires it, rather than making it a policy for every turn).
I’ve not been able to reach a clean conclusion as to whether this is a legitimate use case without requiring a storage-permitted account and separate storage rates (I am uncertain what they are today, since the information isn’t searchable online. I could only find the old announcements, and it seems inadequate to pay rates like those just because the harness doesn’t discard the web search results at the end of each turn. And the results aren’t used for anything like model training, where a higher rate is undeniably justifiable).
Could I please ask whether this use case is permissible at the normal API rates available to everyone without signing up for any specific storage-permitted accounts? If it does require a storage-permitted account after all - is my read that to comply without registering a storage-permitted account, the harness would need to discard the web search results at the end of each turn?
Could I please inquire what the storage-permitted rates are currently?
I would also like to ask about an Enterprise account and ZDR agreements. Are there minimum spend requirements and higher-rate-than-standard requirements for eligibility for ZDR? Or can any company reach that agreement at the standard rates by contacting Brave Search sales?
Thanks
LK