fix: copilot by using bucketId and adjust agent models

frontend/lib/chat/Readme.md

@@ -268,7 +268,8 @@ const mcpClient = await createMCPClient({
 
 # Model Configuration
 
-Currently using: `us.anthropic.claude-sonnet-4-20250514-v1:0` via AWS Bedrock
+Currently using: `us.anthropic.claude-sonnet-4-20250514-v1:0` via AWS Bedrock for Router and Pipe Agent.
+Currently using: `us.anthropic.claude-opus-4-6-v1` via AWS Bedrock for Auditor and TextToSql Agent.
 
 ## Data Flow Example
 

frontend/lib/chat/chart/generator.ts

@@ -43,7 +43,7 @@ const chartColors = {
     lines: [lfxColors.positive[500], lfxColors.negative[500], lfxColors.brand[300]],
   },
 };
-const model = bedrock('us.anthropic.claude-sonnet-4-20250514-v1:0');
+const model = bedrock('us.anthropic.claude-opus-4-6-v1');
 
 export async function generateChartConfig(
   results: Result[],

frontend/lib/chat/data-copilot.ts

@@ -5,6 +5,7 @@ import { createAmazonBedrock } from '@ai-sdk/amazon-bedrock';
 import { experimental_createMCPClient as createMCPClient, type LanguageModelV1 } from 'ai';
 import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js';
 import type { Pool } from 'pg';
+import { ofetch } from 'ofetch';
 import type { ChatResponse, IChatResponseDb } from '../../server/repo/chat.repo';
 import { ChatRepository } from '../../server/repo/chat.repo';
 
@@ -56,14 +57,23 @@ export class DataCopilot {
   /** Human-readable overview of tools for router agent decision making */
   private toolsOverview: string = '';
 
+  /** Tinybird bucketId for the current project — fetched once and reused across all pipe calls */
+  private bucketId: number | null = null;
+
   /** Tinybird MCP server URL */
   private tbMcpUrl: string = '';
 
-  /** Amazon Bedrock language model instance */
-  private model: LanguageModelV1;
+  /** Amazon Bedrock language model instance for routing and auditing (Sonnet) */
+  private sonnetModel: LanguageModelV1;
+
+  /** Amazon Bedrock language model instance for text-to-SQL, pipe, and chart agents (Opus) */
+  private opusModel: LanguageModelV1;
+
+  /** Bedrock model identifier for general agents */
+  private readonly BEDROCK_SONNET_MODEL_ID = 'us.anthropic.claude-sonnet-4-20250514-v1:0';
 
-  /** Bedrock model identifier */
-  private readonly BEDROCK_MODEL_ID = 'us.anthropic.claude-sonnet-4-20250514-v1:0';
+  /** Bedrock model identifier for text-to-SQL and pipe agent */
+  private readonly BEDROCK_OPUS_MODEL_ID = 'us.anthropic.claude-opus-4-6-v1';
 
   /** Maximum number of auditor retry attempts */
   private readonly MAX_AUDITOR_RETRIES = 1;
@@ -72,7 +82,8 @@ export class DataCopilot {
   private readonly MAX_SQL_RETRIES = 2;
 
   constructor() {
-    this.model = bedrock(this.BEDROCK_MODEL_ID);
+    this.sonnetModel = bedrock(this.BEDROCK_SONNET_MODEL_ID);
+    this.opusModel = bedrock(this.BEDROCK_OPUS_MODEL_ID);
     this.tbMcpUrl = `https://mcp.tinybird.co?token=${process.env.NUXT_INSIGHTS_DATA_COPILOT_TINYBIRD_TOKEN}&host=${process.env.NUXT_TINYBIRD_BASE_URL}`;
   }
 
@@ -88,10 +99,43 @@ export class DataCopilot {
       }),
     });
 
-    this.tbTools = await this.mcpClient.tools({});
+    const allTools = await this.mcpClient.tools({});
+
+    // Filter out tools with empty descriptions — Bedrock rejects them with a validation error
+    this.tbTools = Object.fromEntries(
+      Object.entries(allTools).filter(([_, tool]: [string, any]) => {
+        const description =
+          (tool?.description as string) || (tool?.meta?.description as string) || '';
+        return description.trim().length > 0;
+      }),
+    );
+
     this.buildToolsOverview();
   }
 
+  /**
+   * Fetch and store the Tinybird bucketId for a project.
+   * Uses ofetch directly to stay outside the Nuxt server context (no useStorage/createError).
+   */
+  private async fetchBucketId(project: string): Promise<void> {
+    if (!project) return;
+    const tinybirdBaseUrl =
+      process.env.NUXT_TINYBIRD_BASE_URL || 'https://api.us-west-2.aws.tinybird.co';
+    const tinybirdToken = process.env.NUXT_INSIGHTS_DATA_COPILOT_TINYBIRD_TOKEN;
+    if (!tinybirdToken) return;
+    try {
+      const response = await ofetch(
+        `${tinybirdBaseUrl}/v0/pipes/project_buckets.json?project=${encodeURIComponent(project)}`,
+        { headers: { Authorization: `Bearer ${tinybirdToken}` }, timeout: 10_000 },
+      );
+      this.bucketId = response.data?.[0]?.bucketId ?? null;
+      console.warn(`🪣 [DataCopilot] bucketId for "${project}": ${this.bucketId}`);
+    } catch (error) {
+      console.error(`[DataCopilot] Failed to fetch bucketId for "${project}":`, error);
+      this.bucketId = null;
+    }
+  }
+
   /**
    * Build human-readable overview of available tools for the router agent
    */
@@ -170,10 +214,20 @@ export class DataCopilot {
     instructions?: string,
   ): Promise<void> {
     const chatRepo = new ChatRepository(insightsDbPool);
+
+    let model: string | undefined = this.BEDROCK_SONNET_MODEL_ID;
+
+    if (agent === 'EXECUTE_INSTRUCTIONS') {
+      model = undefined;
+      // For now, we use the Opus model for text-to-SQL and auditor.
+      // The model is currently too slow for both the pipe and router agents.
+    } else if (agent === 'TEXT_TO_SQL' || agent === 'AUDITOR') {
+      model = this.BEDROCK_OPUS_MODEL_ID;
+    }
     await chatRepo.saveAgentStep({
       chatResponseId,
       agent,
-      model: agent === 'EXECUTE_INSTRUCTIONS' ? undefined : this.BEDROCK_MODEL_ID,
+      model,
       response,
       inputTokens: response?.usage?.promptTokens || 0,
       outputTokens: response?.usage?.completionTokens || 0,
@@ -207,7 +261,7 @@ export class DataCopilot {
   }: Omit<RouterAgentInput, 'toolsOverview' | 'model' | 'tools'>) {
     const agent = new RouterAgent();
     return agent.execute({
-      model: this.model,
+      model: this.sonnetModel,
       messages,
       tools: this.tbTools,
       toolsOverview: this.toolsOverview,
@@ -248,7 +302,7 @@ export class DataCopilot {
 
     const agent = new TextToSqlAgent();
     return agent.execute({
-      model: this.model,
+      model: this.opusModel,
       messages,
       tools: followUpTools,
       date,
@@ -288,12 +342,21 @@ export class DataCopilot {
     const followUpTools: Record<string, unknown> = {};
     for (const toolName of toolNames) {
       if (this.tbTools[toolName]) {
-        followUpTools[toolName] = this.tbTools[toolName];
+        const tool = this.tbTools[toolName] as any;
+        // Wrap execute to inject bucketId into every MCP tool call during planning
+        followUpTools[toolName] =
+          this.bucketId !== null && this.tbTools[toolName]?.execute
+            ? {
+                ...this.tbTools[toolName],
+                execute: async (params: any) =>
+                  tool.execute({ bucketId: this.bucketId, ...params }),
+              }
+            : tool;
       }
     }
     const agent = new PipeAgent();
     return agent.execute({
-      model: this.model,
+      model: this.sonnetModel,
       messages,
       tools: followUpTools,
       date,
@@ -329,7 +392,7 @@ export class DataCopilot {
     const dataSummary = generateDataSummary(data);
     const agent = new AuditorAgent();
     return agent.execute({
-      model: this.model,
+      model: this.opusModel,
       messages,
       originalQuestion,
       reformulatedQuestion,
@@ -606,7 +669,7 @@ export class DataCopilot {
       }
 
       // Prepare for retry - add feedback to messages and loop
-      previousFeedback = auditorResult.feedback_to_router;
+      previousFeedback = auditorResult.feedback_to_router || undefined;
       attemptNumber++;
 
       dataStream.writeData({
@@ -742,6 +805,12 @@ export class DataCopilot {
     const parametersString = JSON.stringify(parameters || {});
     const date = new Date().toISOString().slice(0, 10);
 
+    // Fetch bucketId once upfront — required by all Tinybird pipes for data partitioning
+    const project = (parameters as any)?.project;
+    if (project) {
+      await this.fetchBucketId(project);
+    }
+
     // Build messages from conversation history
     const { messages, previousWasClarification } = await this.buildMessagesFromConversation(
       currentQuestion,
@@ -768,7 +837,7 @@ export class DataCopilot {
       userPrompt: currentQuestion,
       inputTokens: 0,
       outputTokens: 0,
-      model: this.BEDROCK_MODEL_ID,
+      model: this.BEDROCK_SONNET_MODEL_ID,
       conversationId: conversationId || '',
       routerResponse: RouterDecisionAction.STOP,
       routerReason: '',
@@ -866,7 +935,7 @@ export class DataCopilot {
         routerReason: routerOutput.reasoning,
         pipeInstructions: undefined,
         sqlQuery: undefined,
-        model: this.BEDROCK_MODEL_ID,
+        model: this.BEDROCK_SONNET_MODEL_ID,
         conversationId: conversationId,
       },
       insightsDbPool,
@@ -901,7 +970,7 @@ export class DataCopilot {
         clarificationQuestion: routerOutput.clarification_question || undefined,
         pipeInstructions: undefined,
         sqlQuery: undefined,
-        model: this.BEDROCK_MODEL_ID,
+        model: this.BEDROCK_SONNET_MODEL_ID,
         conversationId: conversationId,
       },
       insightsDbPool,
@@ -1186,7 +1255,7 @@ export class DataCopilot {
     // Execute the pipes according to the instructions and combine results (don't stream data yet, auditor will do it)
     const pipeExecutionStart = Date.now();
     try {
-      const combinedData = await executePipeInstructions(pipeOutput.instructions);
+      const combinedData = await executePipeInstructions(pipeOutput.instructions, this.bucketId);
       const pipeExecutionTime = (Date.now() - pipeExecutionStart) / 1000;
 
       // Track successful pipe execution step
@@ -1246,7 +1315,7 @@ export class DataCopilot {
         routerReason: routerOutput.reasoning,
         pipeInstructions,
         sqlQuery,
-        model: this.BEDROCK_MODEL_ID,
+        model: this.BEDROCK_SONNET_MODEL_ID,
         conversationId: conversationId,
       },
       insightsDbPool,

frontend/lib/chat/instructions.ts

@@ -44,14 +44,19 @@ async function executeTinybirdPipe(
 }
 
 // Function to execute pipe instructions and combine results
-export async function executePipeInstructions(instructions: PipeInstructions): Promise<any[]> {
+export async function executePipeInstructions(
+  instructions: PipeInstructions,
+  bucketId?: number | null,
+): Promise<any[]> {
   // Execute the pipes according to the instructions
   const pipeResults: Record<string, any[]> = {};
 
   // Execute each pipe with its inputs using TinyBird API
   for (const pipeInstruction of instructions.pipes) {
     try {
-      const result = await executeTinybirdPipe(pipeInstruction.name, pipeInstruction.inputs);
+      const inputs =
+        bucketId !== null ? { bucketId, ...pipeInstruction.inputs } : pipeInstruction.inputs;
+      const result = await executeTinybirdPipe(pipeInstruction.name, inputs);
       pipeResults[pipeInstruction.id] = result;
     } catch (error) {
       console.error(`Error executing pipe ${pipeInstruction.name}:`, error);

frontend/lib/chat/prompts/auditor.ts

@@ -36,6 +36,8 @@ export const auditorPrompt = (
     })
     .join('\n');
 
+  const dataSection = `\n## TOP ROWS (first ${dataSummary.topRows.length} of ${dataSummary.rowCount} — use these actual values in your summary)\n\`\`\`json\n${JSON.stringify(dataSummary.topRows, null, 2)}\n\`\`\``;
+
   return `You are an Auditor agent that validates whether retrieved data can answer the user's question.
 
 ## USER'S QUESTION
@@ -47,6 +49,7 @@ ${reformulatedQuestion}
 ## DATA SUMMARY
 **Total Rows:** ${dataSummary.rowCount}
 **Columns:** ${dataSummary.columns.join(', ')}
+**Top Rows:** ${dataSection}
 
 **Column Statistics:**
 ${statsFormatted}
@@ -121,12 +124,12 @@ Make a **BINARY decision**: Can this data answer the user's question?
 
 **IF is_valid = true:**
 - Set \`is_valid: true\`
-- Write a brief \`summary\` (2-3 sentences) for the user:
-  - What the data shows
-  - Key findings based on statistics
-  - Direct answer to their question
-  - Example: "Commit activity in 2024 ranged from 0 to 453 per day across 12 companies, 
-  with an average of 87 commits daily."
+- Write a conversational \`summary\` (1-3 sentences) for the user:
+  - If possible, write a summary that directly answers the user's question.
+    - You SHOULD reference actual values from the TOP ROWS data above — name the specific country, person, organization, etc. NEVER guess, infer, or use external knowledge.
+    - Unknown / null / placeholder entries: If a top row has a null, empty, "Unknown", or placeholder value (e.g. country code "XX", name "null") in a label column, do NOT treat it as a real result. Explain it represents unattributed or anonymous data (e.g. "contributions where the country of origin is unknown"), then identify and state the top row with a real value as the actual answer.
+      - ✅ Example: "The top contributor country is the United States with 670 contributors. Note: 5,882 contributions have no country attribution and are listed separately as 'Unknown'."
+  - Write summary in plain text, not markdown.
 
 **IF is_valid = false:**
 - Set \`is_valid: false\`

frontend/lib/chat/types.ts

@@ -120,9 +120,10 @@ export const auditorOutputSchema = z.object({
   reasoning: z.string().describe('2-3 sentences explaining the validation decision'),
   feedback_to_router: z
     .string()
+    .nullable()
     .optional()
     .describe('If invalid, specific guidance for router to fix the issue'),
-  summary: z.string().optional().describe('If valid, user-friendly summary of findings'),
+  summary: z.string().nullable().optional().describe('If valid, user-friendly summary of findings'),
 });
 
 // TypeScript types for agent outputs

frontend/lib/chat/utils/data-summary.ts

@@ -5,6 +5,7 @@ export interface DataSummary {
   rowCount: number;
   columns: string[];
   columnStats: Record<string, ColumnStats>;
+  topRows: Record<string, unknown>[];
 }
 
 export interface ColumnStats {
@@ -32,8 +33,8 @@ export interface ColumnStats {
 
 /**
  * Generate statistical summary of dataset
- * Token-efficient: ~400-500 tokens for typical dataset
- * No raw data samples sent to LLM - only statistics
+ * Token-efficient: ~1500-200 tokens for typical dataset
+ * Top rows of raw data sent to LLM + statistics
  *
  * @param data - Array of data rows
  * @returns Statistical summary optimized for auditor validation
@@ -44,6 +45,7 @@ export function generateDataSummary<T extends Record<string, unknown>>(data: T[]
       rowCount: 0,
       columns: [],
       columnStats: {},
+      topRows: [],
     };
   }
 
@@ -127,9 +129,13 @@ export function generateDataSummary<T extends Record<string, unknown>>(data: T[]
     columnStats[col] = stats;
   }
 
+  const rows = data as Record<string, unknown>[];
   return {
     rowCount: data.length,
     columns,
     columnStats,
+    // Ideally the entire response would be available to the auditor. But that would be too costly.
+    // TODO: Explore a better way to have a proper summary of the data that answers the user's question directly.
+    topRows: rows.slice(0, 3),
   };
 }

frontend/server/api/chat/chart.ts

@@ -4,6 +4,8 @@ import { Pool } from 'pg';
 import { generateChartConfig, modifyChartConfig } from '../../../lib/chat/chart/generator';
 import { ChatRepository } from '../../repo/chat.repo';
 import { Result, Config, DataMapping } from '../../../lib/chat/chart/types';
+import { getBucketIdForProject } from '../../data/tinybird/bucket-cache';
+import { fetchFromTinybird } from '../../data/tinybird/tinybird';
 import { PipeInstructions } from '~~/lib/chat/types';
 
 export const maxDuration = 30;
@@ -59,7 +61,9 @@ export default defineEventHandler(async (event): Promise<ChartConfigResponse | E
       const { executePipeInstructions } = await import('../../../lib/chat/instructions');
 
       try {
-        const executedResults = await executePipeInstructions(pipeInstructions);
+        const project = pipeInstructions.pipes[0]?.inputs?.project as string | undefined;
+        const bucketId = project ? await getBucketIdForProject(project, fetchFromTinybird) : null;
+        const executedResults = await executePipeInstructions(pipeInstructions, bucketId);
 
         if (!userQuery) {
           return createError({