' j { dZddlZddlZddlZddlZddlZddlmZ ddl m Z ddl m Z ddlmZddlmZmZmZddlmZdd lmZdd lmZd Zd d ddddZdZdeededZedzezZeejdeddde Z ddde!de!dzde"dzde#e$e!e!fdzd e%e$dze!ff d!Z&d"e!d e%e!e#e$e!e!fffd#Z'e j(d$e e d"e!d e!fd%Z)e j(d$e e d e!fd&Z*e j(d$e e de!d e!fd'Z+e j( d0d$e e de!d)e!d e!fd*Z,e j( d1d$e e d+e!de!d,e!d-e!d.e!d e!fd/Z-dS)2u;Chanakya routing agent — tools for consulting sub-agents.N)logger) RunContext)RequestContext)EnterpriseAgent)TATASTEEL_QUERY_URLUI_AGENT_STREAM_URLget_http_client)search_and_fetch) summarise)make_decision_framework_promptuYou are Enterprise Brain, the central intelligence hub. You have four capabilities: ask_tatasteel, ask_tatasteel_decision_support, ask_ui_agent, and web_search. ━━━ CAPABILITIES ━━━ ask_tatasteel — queries the live construction project database. Use for: contracts, early warnings, compensation events, quotations, project figures, rankings, trends, counts, or any data in the project database. When both primary data and market context are needed, batch this in parallel with web_search. ask_ui_agent — ALWAYS call this after ask_tatasteel, no exceptions. Renders results as styled frontend widgets with charts and narrative. Pass market_context and storyline_hints so the response is fully enriched. web_search — searches the internet for real-time market context. Stores market_context for ask_ui_agent. When both primary data and market context are needed, batch this in parallel with ask_tatasteel. ask_tatasteel_decision_support — runs a follow-up Tata Steel analysis query aimed at action, prioritisation, delay reduction, cost reduction, tradeoff, or contractor focus. Stores decision-support data for ask_ui_agent. ━━━ WORKFLOW FOR DATABASE QUESTIONS ━━━ STEP 1 — INTENT ANALYSIS (think first, call no tools yet): a. Decode business terms in the question — what is the user really asking? Translate jargon into the underlying data question. b. Classify the question type: comparison, ranking, trend, proportion, single KPI. c. Identify the underlying decision the user needs to make — not just the data they asked for. Surface questions often mask prioritisation needs. Ask: what does this person need to act on? d. Map the question to the closest Decision Context (DC-01 through DC-09) from the Decision-Enabling Framework. Then generate exactly 5 follow-up questions anchored to that DC: - 2 from the open decision options of the matched DC — what the persona still needs to decide after receiving this answer - 2 from the required data signals of the matched DC that the primary query did not yet surface — data gaps that would complete the decision picture - 1 bridging question pointing to the adjacent DC the persona should address once the current decision is resolved All 5 must be answerable from the database and must advance the persona's mandate. Never frame NCE auto-approval as a cost or schedule risk in follow-up questions — cost and schedule impact comes only from accepted quotations (change_to_prices, change_to_days); auto-approval affects the PM's reject right, not the budget. CRITICAL — follow-up questions are shown verbatim to end users. They MUST be written in plain business language. Never include raw SQL column names (e.g. change_to_prices, change_to_days, reply_due_date, record_number, vendor, nce_number). Translate every technical term: "change_to_prices" → "cost impact", "change_to_days" → "time impact", "reply_due_date" → "response deadline", "record_number" → "reference", "vendor" → "contractor". If a question reads like a SQL query, rewrite it. e. Compose a dc_mapping_explanation (3–6 lines) that states: - Which DC was matched and why (e.g. "Mapped to DC-08 — Finance budget coverage, because the question asks how to prevent budget overruns from a Finance lens") - For each of the 5 follow-up questions, one sentence explaining which decision option or required data signal from the matched DC it comes from Pass this as dc_mapping_explanation when calling ask_ui_agent. It is rendered as a visible section so the user can see how the follow-up questions were derived. STEP 2 — Call ask_tatasteel with the primary question to get data and explanation. This first ask_tatasteel call is the main answer dataset. STEP 3 — In the SAME model turn, batch these two independent tool calls in parallel when both are needed: - ask_tatasteel for the primary answer dataset - web_search for external market / industry context Only call web_search when the question requires context the database cannot provide — industry benchmarks, market rates, contractor reputation, or sector news. Skip it when the database alone fully answers the question. STEP 4 — After ask_tatasteel and web_search complete, decide whether actionability needs extra support: - Call ask_tatasteel_decision_support zero, one, or more times with tightly-scoped follow-up analytical questions. - Do this AFTER reviewing the primary answer and any market context. STEP 5 — Call ask_ui_agent with ALL enrichment: • question: the original user question (verbatim) • market_context: optional; leave blank if web_search already stored it • storyline_hints: the 5 follow-up questions from Step 1d, pipe-delimited • dc_mapping_explanation: the DC mapping explanation from Step 1e ━━━ VISUALIZATION FOLLOW-UPS ━━━ (when user asks to change/switch/modify the chart or visualization) The previous data is already cached — do NOT call ask_tatasteel or web_search again. Call ask_ui_agent DIRECTLY with a visualization_hint that describes: - The chart type currently shown (find [Visualisation shown: ...] in prior reply) - What the user wants, e.g. 'user wants bar chart' Example: 'Currently showing contract-value-orb. User wants a different visualization.' Skipping ask_tatasteel AND web_search is REQUIRED — avoids needless round-trips. CURRENCY CONVERSION REQUESTS (e.g. 'convert to dollars', 'show in euros', 'change to rupees'): The underlying data does not change — do NOT call ask_tatasteel. Call ask_ui_agent DIRECTLY with the user's question. The UI agent will retrieve the previous rows from conversation history and re-render them in the new currency. ━━━ DIRECT ANSWERS (use no tools) ━━━ - Greetings, general conversation, help requests. - Questions about yourself or your capabilities. - Anything answerable from conversation history alone. - General knowledge not requiring project database data. NOTE: 'change the visualization', 'show as a different chart' are NOT direct answers — always use tools for these. CONVERSATION HISTORY: You have full access to prior messages. Use it to answer follow-up questions accurately without re-querying the database. zFocus on delivery cost and timelines as primary drivers of project success. Prioritize interventions that reduce notification cycle time and cost exposure. zFocus on identifying and mitigating contractual and schedule risks. Prioritize early warnings, compensation events, and potential disputes before they escalate. zFocus on vendor performance metrics and accountability. Prioritize vendor compliance, delivery consistency, and quality benchmarks. zFocus on cash flow optimization and financial planning. Prioritize quotation approvals, payment schedules, and working capital efficiency. zFocus on scope management and change control. Prioritize variation identification, scope creep prevention, and contract alignment. )cost_and_timelinerisk_mitigationvendor_performance cash_flow scope_controlr utTATA STEEL DEPLOYMENT CONTEXT: This Enterprise Brain deployment is for Tata Steel. ━━━ PERSPECTIVE ━━━ u  ━━━ USER CONTEXT ━━━ Users span the full organisational hierarchy — board members, C-suite executives (CTO, CFO, COO, CEO), programme directors, construction management heads, finance leads, commercial managers, contract administrators, and project managers. All are familiar with NEC contract terminology and steel industry operations. NEVER define or explain NCE, EW, CE, NEC, quotation, compensation event, variation, or any standard industry term — treat all domain terminology as shared knowledge. Do not add disclaimers or jargon explanations at any level of seniority. TATA STEEL WEB SEARCH SCOPING: - Use for: industry benchmarks, Tata Steel UK news, NEC contract norms, UK construction cost trends, contractor market intelligence, steel sector news. u ━━━ TATA STEEL INTENT EXAMPLES: - Decode NEC terms such as NCE, EW, and CE into the underlying data question. - Example follow-up lines for Tata Steel questions: Question: 'NCE variance across vendors' Storylines: • 'Which vendors have the highest NCE-to-CE conversion rate?' • 'How has NCE volume trended month-over-month for the top 3 vendors?' • 'What is the average financial value of NCEs per vendor?' • 'Which vendor cluster should be reviewed first to avoid the largest cost exposure?' • 'Which intervention would most likely reduce notification cycle time over the next month?' TATA STEEL STORYLINE FORMAT EXAMPLE: 'Which vendors have highest NCE-to-CE rate? | How has NCE volume trended? | What is average NCE value per vendor? | Which vendor cluster should be reviewed first to avoid the largest cost exposure? | Which intervention would most likely reduce notification cycle time over the next month?' PERSPECTIVE FILTERING RULE: All market context, external trends, and decision-support insights MUST be evaluated against the active PERSPECTIVE and either: a) Filtered OUT if irrelevant to the perspective focus, OR b) Explicitly LINKED to the perspective's primary drivers if included. Do not include external context (market trends, industry benchmarks, labor market) unless it directly informs or accelerates action aligned with the current perspective.  LITELLM_PROVIDER_BASE_URLzchanakya-alphachanakya)api_base instructionsretries logfire_env service_name deps_type)message_history_wirequestion user_emailconversation_idrreturncK t}|t||||dd{V}nD#tj$rYdSt $r&}t jd|dd|fcYd}~Sd}~wwxYw|jdkr dd|jdfS| }| d d kr1| d i}dd | d dfS|dfS)N)rrrhistoryjson)Nz3Could not reach the Tatasteel agent. Is it running?z[ASK_TATASTEEL] HTTP error: {}z"Error contacting Tatasteel agent: zTatasteel agent returned HTTP .statusfailureerrorzTatasteel agent error: messagez unknown error) r postrhttpx ConnectError Exception_logr( status_coder#get) rrrrclientrespexcpayloaderrs D/var/www/html/ai-enterprise-brain/backend/chanakya/rt_agent/agent.py_query_tatasteelr8sq@ ""[[ $(#2/ !          KKKJJJ @@@ 3S999?#?????????@ 3Id6FIIIIIiikkG{{8 ))kk'2&&Tswwy//R/RTTTT B;s!5:A; A;A60A;6A;querycKtjt|d{V}tjt||d{V\}}||fS)N)asyncio to_threadr r )r9rawtextclaimss r7_run_web_searchr@s_!"2E:: : : : : : :C *9eSAAAAAAAALD& <ctxcK|jr |jjnd}||ddddd{Vt|d{V\}}|j||j_||j_|S)u4Search the web and return grounded, source-cited market context bullets. Each bullet is backed by a verbatim quote from the search results. Claims without a verifiable source are excluded automatically. Use the returned text verbatim as market_context — do NOT re-summarise or augment it. Nprogress searchingzReading web sources...typestepr))depsprogress_queueputr@ web_claimsmarket_context_text)rBr9rJr>r?s r7 web_searchrNsGJh+X38+B+BTXN!  *kVn!o!oppppppppp(////////LD& x$'+$ KrAcr|jsdSd|jjpdd|jjpdd|jjpdS)zEReturn the calling user's identity and role from the request context.z6user_email=(unknown), role=(unknown), tenant=(unknown)z user_email=z (unknown)z, role=z , tenant=)rIrroletenant)rBs r7get_my_contextrR sa 8HGG 3ch)8[ 3 3 , 3 3(/0[ 3 3rAc LK|jr |jjnd}|jr |jjnd}|jr|jjr |jjnd}d|d|}t jd||tj}|jr |jj nd}|| dddd d{Vt||||jr |jj nd d{V\}} || S| d d } | d g} | dd} t jdt| t| | |4| dgD]} | | d{V|j|g|j_|jj*||j_|jj| d|j_| d|j_| dpg|j_|jjd|t| | | | t1tj|z dzd| S)a'Query the Tatasteel agent via HTTP /query and return the explanation text. Calls tatasteel synchronously to get SQL results (rows + explanation). The payload is stored in ``ctx.deps.tatasteel_result`` so ``ask_ui_agent`` can pick it up directly without a second tatasteel call. NPM [PERSONA: ] z*[ASK_TATASTEEL] question={!r} persona={!r}rDqueryingzQuerying Tata Steel database...rFrrr explanation(no explanation)rows generated_sqlr*z5[ASK_TATASTEEL] explanation={} chars, rows={}, sql={}progress_stepssql_explanation query_log ask_tatasteel)rHrr[rYr\rows_all duration_ms)rIrrrPstripr/debug_time monotonicrJrKr8rr1lentatasteel_resulttatasteel_primary_result"tatasteel_decision_support_resultscleartatasteel_generated_sqltatasteel_sql_explanationtatasteel_query_logflow_logappendround)rBrrrpersonatagged_question _ts_startrJr5err_msgrYr[sql step_events r7r`r`s),:$$dJ25(Dch..O #DSX]Dsx}}KKMMG8788h88OJ;XwOOO!!IGJh+X38+B+BTXN!  *jUv!w!wxxxxxxxxx-'>AhPSX::D GW ++m-?@@K ;;vr " "D ++or * *CJFKHXHXZ]^bZcZcehiii!!++&6;; 1 1J $$Z00 0 0 0 0 0 0 0 0 x%,I! 8 , 407CH - H 7 = = ? ? ?+2;;+G+G(-4[[9J-K-K*'.{{;'?'?'E2$   # II&  %/"3"3i"?4!GHH" "     rAr*focusc |K|jr |jjnd}|jr |jjnd}|jr|jjr |jjnd}d|d|}t jd|||tj}t||||jr |jj ndd{V\}} || S| dgpg} | dd } | d d } |j|jj ||pd| | | | d | d pgd|jj d||pdt| | | | t!tj|z dzd| S)z@Run a follow-up Tata Steel query used only for decision support.NrTrUrVzF[ASK_TATASTEEL_DECISION_SUPPORT] question={!r} focus={!r} persona={!r}rXr[rYrZr\r*r^r_)rryr[rYr\r^r_ask_tatasteel_decision_supportra)rHrryr[rYr\rbrc)rIrrrPrdr/rerfrgr8rr1rkrqrprhrr) rBrryrrrsrtstartr5rvr[rYrws r7r{r{Us ),:$$dJ25(Dch..O #DSX]Dsx}}KKMMG8788h88OJWYachjqrrr O  E-'>AhPSX::D GW  ;;vr " " (bD++m-?@@K ++or * *C x 3:: ]d& &{{+<== [117R< <       4 ]dII&  %/"3"3e";t!CDD " "  rAvisualization_hintmarket_contextstoryline_hintsdc_mapping_explanationcK|jr+|jjr|jj}n>|jr1|jjr%|jjd}ngddd}|jr |jjnd}|jr |jjnd}|p|dd} |r| d|dn| } |rd|d Dng} |p|jr |jjnd} |d d } |d g|d d| ||| | pd| pd|pd|jr |jj nd|jr"|jj rd|jj Dnd|dpd|dpd|dpdd}|jr |jj nd}|| ddddd{Vt|d g}|d g}tjd|t!j} t%}|dt(|4d{V}|jdkrHdd|jdd}|| |d{Vd|jd cdddd{VSd}|23d{V}|s ||z }d!|vr|d!d"\}}|d#D]v}|d$r] t1j|d%d}|d&d'kr|jrt5|d(pg}|jjpg}|rat9|D]P\}}|d)|d*d|d pd|dkr|ngd+Qn:|jjr.|d)|jj|jj|d+|jj pg}|D]p} | dd}!| d,p| d pd}"| d pg}#| d-pg}$|$rt|$d"z }%t9|$D]r\}&}|d p|"pd}'|!rd.|!d/|' }'|d)|d*d|'pd|&|%kr|#ngd+s | d0rP|"pd}'|!rd.|!d/|' }'|d)| d0d|'pd|#d+r|r9d1|D}(|(r+|d2d!!|(d3|jj"r"|d |jj"d3i|d(|i}|| |d{V|d&d'kr|jr ||j_#e#tH$rYrwxYwxd!|v6 dddd{Vn#1d{VswxYwYn#tJj&$r'dd4dd}|| |d{VYd5StH$rT})tj'd6|)dtQ|)dd}|| |d{Vd7|)cYd})~)Sd})~)wwxYw|jr9|jj#r-t|jj#d(gnd}*g}+|jr,|jj#r |jj#d(g}+tSd8|+Dd},|jU|jj*d9|pd| | pd| |||*|,pd|+tWt!j|z d:zd; d<|*d=|d>S)?uRender the most recent tatasteel query result as a frontend widget. ALWAYS call this after ask_tatasteel — for every database question without exception. The UI agent renders charts when rows are present and formatted text cards when they are not. Never return a raw tatasteel explanation directly. The payload from the last ask_tatasteel call is forwarded automatically — you do not need to pass rows or explanation explicitly. Args: visualization_hint: Optional hint about preferred chart style (e.g. "bar chart", "show as table", "trend line"). Leave empty to let the UI agent pick the best fit automatically. question: The current user question. Pass this verbatim so the UI agent receives the exact wording (especially for follow-up viz requests like "change the chart"). market_context: Bullet-point summary of web search results — industry benchmarks, Tata Steel UK news, NEC norms, market trends. Pass verbatim from your web_search summary (150–300 words). storyline_hints: Pipe-delimited list of 5 follow-up enquiry questions derived from the matched Decision Context (DC-01 through DC-09). dc_mapping_explanation: Plain-text explanation (3–6 lines) stating which DC was matched and why each of the 5 follow-up questions was derived from that DC's decision options or required data signals. This is rendered as a visible "How follow-ups were derived" section in the UI. rr*)r[rYrNrz [Visualization context: ]c^g|]*}||+S)rd).0ss r7 z ask_ui_agent..s-DDDq!''))DDDDrA| currency_codeGBPr[rYcJg|] }d|D!S)c"i|] \}}|dv || S))r\r^r_r)rkvs r7 z+ask_ui_agent...s)jjj$!Qq8i/i/iA/i/i/irA)itemsrps r7rz ask_ui_agent..s?   kj!''))jjj   rAwhat_to_act_onurgencycost_of_inaction)r[rYrrrrr~rdc_explanationr!decision_support_contextrrrrD renderingzPreparing response...rFz/[ASK_UI_AGENT] calling UI agent stream, rows={}POSTr"r$r(zUI agent returned PIPELINE_ERROR)rGr)codezUI agent returned HTTP r%r zdata: rGdonedata sql_queryrw)rGcontentrYr[r^r_uDecision support —  r\cg|]R}|dd*|ddSS)rYr*)r1rdrs r7rz ask_ui_agent..0sb8.8.8.4534553K3K3Q3Q3S3S8.01mR0H0H0N0N0P0P8.8.8.rAdecision_context)rGrzCould not reach the UI agent.z,Could not reach the UI agent. Is it running?z[ASK_UI_AGENT] error: {}zError contacting UI agent: c3K|]G}|ddk|diddVHdS)rG visualizationrr*N)r1)rws r7 zask_ui_agent..WsUgg!aeeFmmWfFfFfy"   ! !&" - -FfFfFfFfggrA ask_ui_agentra) rHr}rr~ storylinesui_request_forwarded input_rowsoutput_widgetsviz_type widgets_allrczVisualization generated with z widget(s) for z data rows.),rIrjcopyrirrr1splitrMrrkrJrKrhr/rerfrgr streamrr0 aiter_text startswithr#loadslistro enumeraterqrmrnrdjoinrL ui_resultr.r,r-r(strnextrprr)-rBr}rr~rrr5rr base_questionfinal_questionparsed_storylinesresolved_market_contextr ui_requestrJ row_count primary_rows _ui_startr2r3r6 sse_bufferchunksse_msglineparsed done_datar_idxentrydecision_supportpacketpacket_questionpacket_explanation packet_rows packet_logslast_idxlog_idxexplds_partsr4 widget_countrrs- r7rrs F xBCH5B(388:: Bch/B(+A.3355 bbAA(+:$$dJ25(Dch..O; J ; ;M  =II4FIIII   DDO11#66DDDD  -bQTQY1a1M1M_aKK77M FB''{{="55" *&19T,408D47HF3800$ x   HG    D     !++&677?4;;y))1T#KK(:;;Ct+J0GJh+X38+B+BTXN!  *kVm!n!noooooooooGKK++,,I;;vr**LJ@)LLL!!Ia3 ""==)<:=NNT )T )T )T )T )T )T )RV#s**#*7^DL\7^7^hxyyC%1,00555555555HT5EHHH T )T )T )T )T )T )T )T )T )T )T )T )T )T )  #'??#4#4M)M)M)M)M)M)M)% ! %'J J...8.>.>vq.I.I+$+MM$$7$7G)G)D#x88F)E!)-1ZQRR-A-AF'-zz&'9'9V'C'C'C48F9K9K9Qr4R4R 47H4P4VTV +4)/>G >R>R-3-3 U090@0@8;[?Bx?a8D >.>.-/-/-/ DjjUW>X>XOAGL]A^A^BDbhblblmzb{b{BD@D,>:@**V:L:L:RPRK:@**[:Q:Q:WUWK/:-3;>{;K;Ka;OFOP[F\F\ 17 17NGU;@99];S;S;oWi;omoD7F5x?oWf?o?oim?o?o?u?u?w?w4=4D4D@KCH99UTVCWCWGK|tOVZbObOb hj F6F657575757 1728O1L1L -37I7OR3B1t;kSb;k;kei;k;k;q;q;s;sD090@0@)v,5,<,>>+JTdee  % $$S)) ) ) ) ) ) ) )=== 333 -s3333s88=MNN  % $$S)) ) ) ) ) ) ) )2S22222222 3?Bhd38K]d3sx)--fb99:::cdL!K x9CH&9h(,,VR88 gg{ggg H  x   ""4"<&5=+$.#* (D& %/"3"3i"?4!GHH " "  _< ^ ^ ^ ^ ^^s#1[%A[[%([>Z?A[M Z%#[% Z3 /[2Z3 3 [[% [[% [![%%2]7 ]7#A ]2,]72]7)r*)r*r*r*r*r*).__doc__r;r#ostimerfr,logururr/ pydantic_airbackend.chanakya.a2a_contextrbackend.chanakya.baserbackend.chanakya.rt_agent._httprrr 'backend.chanakya.rt_agent.web_groundingr (backend.chanakya.rt_agent.web_summariserr -backend.chanakya.tatasteel.decision_frameworkr _CHANAKYA_GENERAL_PROMPT_TATASTEEL_PERSPECTIVES_DEFAULT_PERSPECTIVE_CHANAKYA_TATASTEEL_PROMPT_CHANAKYA_AGENT_PROMPTenvironrrintrdicttupler8r@toolrNrRr`r{rrrAr7rsAA  !!!!!!""""""777777111111eeeeeeeeeeDDDDDD>>>>>>XXXXXXeP#.+* -. **& !!'***X2F:=WW ? Z3 4'      $9= d 4Z  tCH~.5   4$;  DsDc3h4H/H)I  *^4SS&  >2s 9Z793939999x // N #// / ////d !"$ a_a_ N #a_a_a_ a_  a_  a_ a_a_a_a_a_a_rA