Personalization

The maximum context displayed by providers is much higher than the context that LLM can handle effectively, most LLM that display a maximum context of 164k do not perform very well when the context exceeds 60k tokens.

BDS1 is designed to work with LLMs such as Deepseek V3.2, Minimax M2, or GLM 4.6. For these LLMs I recommend not adding too many search functions but you can put approximately two “Three Brained Searches” for each cycle to version x3 (3 cycles).

The more the search progresses, the more the LLM’s context fills up, so it is better to have 2 or 3 search functions in cycles 1, 2 functions in cycles 2 and only 1 functions in cycle 3, so that the LLM is not overloaded at the end.

To add a function, you must install the tool, select it in the BDS1 agent settings, and add the function under : three_smart_searches({“query_1”: “keywords-A1”, “query_2”: “keywords-A2”, “query_3”: “keywords-A3”, “long_query”: “tool-query-A”}).

Add a RAG search

If you have vectorized (embedded) documents to create a RAG, to use it you need a tool that searches your RAG and then, for each cycle, adds the command to execute the RAG tool function below the search functions already present in steps 3, 10 and 17.

For example, if the function is called “find”, for the first cycle, you will have the function to search the internet followed by the “find” function to search the RAG, which will search based on the keywords A1, maybe you can put 3 “find” function for “keywords-A2” and “keywords-A3”.

three_smart_searches({“query_1”: “keywords-A1”, “query_2”: “keywords-A2”, “query_3”: “keywords-A3”, “long_query”: “tool-query-A”}) find({“query”: “keywords-A1”})