ٟiUdZddlZddlZddlZddlZddlmZmZmZm Z ejddZ ejddZ dZ d Zd Zejd d Zeszd Zej$j'er]ee5ZeD]GZej/dsej1dddj3j3dZnddddadZdedeefdZ d deededede edeeeeff dZ d dededede edeeeeff dZ!d!dededefdZ"dZ#e$dk(re#yy#1swYrxYw)"a Genesis RAG Query Engine ======================== Semantic search across the bloodstream knowledge base. Embeds a query with gemini-embedding-001 (3072d), searches Qdrant genesis_memories collection, optionally enriches with PostgreSQL metadata. Usage: python3 -m core.rag_query "How does Telnyx assistant creation work?" python3 -m core.rag_query --top 10 "GHL pipeline setup" python3 -m core.rag_query --json "SubAIVA provisioning" N)ListDictAnyOptional QDRANT_URLz/https://qdrant-b3knu-u50607.vm.elestio.app:6333QDRANT_API_KEY@7b74e6621bd0e6650789f6662bca4cbf4143d3d1d710a0002b3b563973ca6876genesis_memorieszgemini-embedding-001i GEMINI_API_KEYz(/mnt/e/genesis-system/config/secrets.envzGEMINI_API_KEY==z'"cRtddlm}|jtatS)Nr)genai)api_key)_clientgooglerClientr ) google_genais '/mnt/e/genesis-system/core/rag_query.py_get_genai_clientr6s"0%%n%= Ntextreturnct}|jjt|}t |j dj S)z2Embed a query string using Gemini embedding model.)modelcontentsr)rmodels embed_content EMBED_MODELlist embeddingsvalues)rclientresults r embed_queryr&>sG  F ]] ( ()F !!!$++ ,,r query_vectortop_kscore_threshold type_filtercddlm}ddlm}m}m}|t t}d} |r'||d||jg } |jt|||| d } g} | jD]} | jxsi} | jt| jt!| j"d | j%d d| j%dd| j%ddxs| j%dd| j%ddxs| j%dd| j%dg| j%dd| j%ddd | S)z*Search Qdrant genesis_memories collection.r) QdrantClient)FilterFieldCondition MatchValue)urlrNtype)value)keymatch)mustT)collection_namequerylimitr) query_filter with_payloadtitler source_urlsourcercontenttagsplatform confidence) idscorer<r1r>r?r@rArB) qdrant_clientr,qdrant_client.modelsr-r.r/rrupper query_points COLLECTIONpointspayloadappendstrrCroundrDget)r'r(r)r*r,r-r.r/r$r9resultshitspointrKs r search_qdrantrSKsP+GG j. AFLj{7H7H7J&K !!"'! "G D --%2 %((mu{{A. Wb1 FB/!++lB7T7;;xQS;T";;vr2Pgkk)R6P FB/#KK B7%kk,:   Krquestionc6t|}t||||S)ai Full RAG query: embed question -> search Qdrant -> return results. Args: question: Natural language query top_k: Number of results to return score_threshold: Minimum similarity score (0-1) type_filter: Optional type filter (e.g., "PRODUCT_SPEC") Returns: List of matching knowledge entries with scores )r'r(r)r*)r&rS)rTr(r)r*vectors r rag_queryrWs'$ "F '  rc bt||}|sydt|dg}t|dD]p\}}|jd|d|dd |d d |d d |jd|d|ddd}|jd||jdrdj |S)z RAG query that returns a formatted context string ready for LLM injection. This is the function agents should call to get relevant knowledge before generating a response. r(z,[No relevant knowledge found in bloodstream]z=== BLOODSTREAM KNOWLEDGE (z matches) === r[] r<z (score: rDz, type: r1)z Source: r>r?Niz r  )rWlen enumeraterLjoin)rTr(rPlinesirr?s r rag_contextrds.G =*3w<. H IE'1%1 q2aj\1W:,hqykQRST |AhK=12I,t$ tG9%& R  99U rc |tjd}|jdd|jdtdd |jd td d |jd dd|jddd|jddd|j }|j r+tt|j|jyt|j|j|j|j}|jr&ttj|dt y|s tdytddtd|jtdt#|d t$d!t&tddt)|d"D]\}}td#|d$|d%td&|d'd(|d)d*|d+|d,dd-}td.|d/|d0r#td1d2j+|d0ddty)3NzGenesis RAG Query Engine) descriptionr7z Search query)helpz--topzNumber of results)r1defaultrgz --threshold333333?z Min scorez--typer*zFilter by type)destrgz--json store_truez JSON output)actionrgz --contextzLLM context formatrY)r(r)r*)indentrizNo results found.r]zF======================================================================z RAG QUERY: z Results: z | Collection: z | Dim: rz [r[r<z Score: rDz | Type: r1z | Source: r>r?z z...r@z Tags: z, )argparseArgumentParser add_argumentintfloat parse_argscontextprintrdr7toprW thresholdr*jsondumpsrMr^rI VECTOR_DIMr_r`)parserargsrPrbrcr?s rmainrs  $ $1K LF n5 c1;NO  E3[Q };KL MJ  L?ST    D || k$**DHH56 hh$$ G yy djjC89  !" Bvh- M$** &' KG ~_ZL UV VHB-'1%1 A3b7 %& aj\QvYK}Qx[MZ[I,t$ wis#$ V9 L1V9Ra=!9 :; < r__main__)rhrjN)rh)%__doc__ossysr{rqtypingrrrrgetenvrrrIr r}r secrets_pathpathexistsopenfline startswithsplitstriprrrMrur&rtrSrWrdr__name__rrrs  ,, RYYC F $  +R0=L ww~~l# ,  1 ??#45%)ZZQ%7%:%@%@%B%H%H%ON   -c-d5k- !% 0u+0 00# 0  $sCx. 0p !%  #   $sCx. 6#c#2)X zFq  sE5EE