@duckcodeailabs/dql-cli 1.8.13 → 1.9.1

package/dist/local-runtime.js

@@ -19,7 +19,7 @@ import { fetchLatestPublishedDqlVersion, resolveDqlRuntimeVersionStatus } from '
 import { resolveRetrievalHealthStatus } from './retrieval-health.js';
 import { createDqlAgentProviderRunner, resolveAgentFollowUpContext } from './llm/providers/dql-agent-provider.js';
 import { listRemoteMcpSettings, saveRemoteMcpSettings } from './llm/mcp-config.js';
-import { ClaudeProvider, ConversationStore, advanceThreadState, buildConversationSnapshot, recallRelevantTurns, GeminiProvider, MemoryStore, OllamaProvider, OpenAIProvider, buildBlockBusinessFingerprint, buildBlockSqlFingerprints, buildAnalysisQuestionPlan, aggregationIntegrityIssuesForSql, buildLocalContextPack, toAgentRetrievalEvidence, defaultConversationPath, defaultMemoryPath, ensureDefaultMemoryFiles, ensureAgentProjectReady, isAgentProjectIndexReady, currentMetadataFingerprint, ensureMetadataCatalogFresh, readIndexedDomainKnowledge, readIndexedKnowledge360, propose, proposePlan, recordCorrectionTrace, emitCorrectionEvalCase, mineJoinPatterns, reviewHint, AgentRunEngine, SqliteAgentRunStore, defaultAgentRunGates, createLlmAgentRunPlanner, createHybridRouter, computeResultStats, buildDeterministicDashboardStory, synthesizeAnswer, streamOrGenerate, narrateResult, buildProposePreview, buildFromPrompt, defaultAgentRunStorePath, defaultAgentRunSqlitePath, resolveLocalOwner, resolveProposeConfig, recordQueryRun, recordRuntimeSchemaSnapshot, latestRuntimeSchemaSnapshotForProject, loadSkills, migrateLegacySkills, configuredSkillsPath, skillsDir, draftDomainSkillBootstrap, buildDomainSkillBootstrapPrompt, mergeDomainSkillBootstrapEnrichment, writeSkill, deleteSkill, deriveGeneratedDraftSlug, reindexProject, invalidateAgentProjectState, recordAgentRuntimeVersion, resolveDomainContextEnvelope, defaultKgPath, planAppFromPrompt, KGStore, planResearch, loadSemanticMetrics, cascadeTraceToEvidenceRouteSteps, createCascadeAnswerResult, createCascadeTrace, routeReasoningEffort, routeForCascadeAnswerTier, clampReasoningEffort, bumpReasoningEffort, resolveThinkingMode, coerceThinkingMode, upsertGeneratedDqlArtifactDraft, loadAgentSemanticLayer, } from '@duckcodeailabs/dql-agent';
+import { ClaudeProvider, ConversationStore, advanceThreadState, buildConversationSnapshot, recallRelevantTurns, GeminiProvider, MemoryStore, OllamaProvider, OpenAIProvider, buildBlockBusinessFingerprint, buildBlockSqlFingerprints, buildAnalysisQuestionPlan, aggregationIntegrityIssuesForSql, buildLocalContextPack, toAgentRetrievalEvidence, defaultConversationPath, defaultMemoryPath, ensureDefaultMemoryFiles, ensureAgentProjectReady, isAgentProjectIndexReady, currentMetadataFingerprint, ensureMetadataCatalogFresh, readIndexedDomainKnowledge, readIndexedKnowledge360, compactSemanticRuntimeFailure, propose, proposePlan, recordCorrectionTrace, emitCorrectionEvalCase, mineJoinPatterns, reviewHint, AgentRunEngine, SqliteAgentRunStore, defaultAgentRunGates, createLlmAgentRunPlanner, createHybridRouter, computeResultStats, buildDeterministicDashboardStory, synthesizeAnswer, streamOrGenerate, narrateResult, buildProposePreview, buildFromPrompt, defaultAgentRunStorePath, defaultAgentRunSqlitePath, resolveLocalOwner, resolveProposeConfig, recordQueryRun, recordRuntimeSchemaSnapshot, latestRuntimeSchemaSnapshotForProject, loadSkills, migrateLegacySkills, configuredSkillsPath, skillsDir, draftDomainSkillBootstrap, buildDomainSkillBootstrapPrompt, mergeDomainSkillBootstrapEnrichment, writeSkill, deleteSkill, deriveGeneratedDraftSlug, reindexProject, invalidateAgentProjectState, recordAgentRuntimeVersion, resolveDomainContextEnvelope, defaultKgPath, planAppFromPrompt, KGStore, planResearch, loadSemanticMetrics, cascadeTraceToEvidenceRouteSteps, createCascadeAnswerResult, createCascadeTrace, routeReasoningEffort, routeForCascadeAnswerTier, clampReasoningEffort, bumpReasoningEffort, resolveThinkingMode, coerceThinkingMode, upsertGeneratedDqlArtifactDraft, loadAgentSemanticLayer, } from '@duckcodeailabs/dql-agent';
 import { gatherProposeEnrichment } from './propose-enrich.js';
 import { handleAppsApi, proposeAppAiBuild, recommendVisualization } from './apps-api.js';
 import { getActiveProvider, getEffectiveProviderConfig, isProviderSettingsId, listProviderSettings, saveProviderSettings, } from './settings/provider-settings.js';
@@ -33,8 +33,8 @@ import { buildSemanticObjectDetail, buildSemanticTree, computeSyncDiff, loadSema
 import { clearBlockStudioImportSessions, candidateToDqlSource, createBlockStudioImportSession, deleteBlockStudioImportSession, listBlockStudioImportSessions, loadBlockStudioImportSession, readBlockStudioImportCandidate, parameterizeSqlForDqlImport, updateBlockStudioImportCandidate, writeBlockStudioImportSession, writeBlockStudioImportCandidate, } from './block-studio-import.js';
 import { MetricFlowUnavailableError, hasDbtSemanticManifest, } from "./metricflow.js";
 import { ManagedMetricFlowInstaller, isMetricFlowWarehouseAdapter, metricFlowAdapterForDriver, } from './metricflow-installer.js';
-import { compileSemanticRuntimeQuery, getSemanticRuntimeStatus, isSemanticRuntimeError, listRuntimeCompatibleDimensions, semanticMetricExecutionCapability as runtimeMetricExecutionCapability, SemanticRuntimeRequiredError, testSemanticRuntimeDraft, } from './semantic-runtime.js';
-import { getSemanticRuntimeSettings, saveTestedSemanticRuntimeSettings, } from './semantic-runtime-settings.js';
+import { compileSemanticRuntimeQuery, describeRuntimeCompatibility, getSemanticRuntimeStatus, isSemanticRuntimeError, semanticMetricExecutionCapability as runtimeMetricExecutionCapability, SemanticRuntimeRequiredError, testSemanticRuntimeDraft, } from './semantic-runtime.js';
+import { getSemanticRuntimeSettings, saveTestedSemanticRuntimeSettings, saveSemanticRuntimePreference, } from './semantic-runtime-settings.js';
 import { NotebookDatasetWorkspace, } from "./notebook-datasets.js";
 import { prepareBlockInvocation } from './block-invocation.js';
 const NOTEBOOK_EXECUTE_PREVIEW_ROW_LIMIT = 500;
@@ -204,15 +204,54 @@ function agentColumnDisplayFormats(projectRoot, columns) {
  * no env override. The defaults must fit the transport's physics. */
 const AGENT_CLI_LOOKUP_DEADLINE_MS = 150_000;
 const AGENT_CLI_RESEARCH_DEADLINE_MS = 300_000;
+/** High reasoning effort makes every LLM call materially slower, so the same
+ * multi-step build that fits a budget at medium effort overruns it at high. The
+ * budget must scale with effort or "High · thorough" reliably times out. */
+function reasoningEffortBudgetMultiplier(effort) {
+    return effort === 'high' ? 1.8 : effort === 'low' ? 0.9 : 1;
+}
+/**
+ * A ranking/breakdown/filtered question needs LLM SQL generation (the grain
+ * dimension usually lives on a joined model the semantic layer can't compose),
+ * and that multi-call build does NOT fit the bare lookup budget on a slow or
+ * high-effort provider — it dead-ends at the deadline. It is not "research", but
+ * it needs research-sized time to finish. Detected from the requested SHAPE (a
+ * top-N, a member filter, a breakdown dimension, or a time window) rather than a
+ * recognized metric word, since office questions like "top bcm customers from
+ * the south region" name no metric the planner knows yet. A bare scalar
+ * ("total revenue") has none of these and stays on the short budget.
+ */
+function isAnalyticalBuildQuestion(plan) {
+    if (!plan.needsGeneratedSql)
+        return false;
+    return plan.requestedShape.dimensions.length > 0
+        || plan.dimensionTerms.length > 0
+        || plan.requestedShape.filters.length > 0
+        || plan.timeTerms.length > 0
+        || Boolean(plan.requestedShape.topN);
+}
 export function agentRunDeadlineMs(request, env = process.env, activeProviderId) {
     const cliProvider = activeProviderId === 'claude-code' || activeProviderId === 'codex';
     const lookupDeadline = resolveAgentDeadlineMs('DQL_AGENT_LOOKUP_DEADLINE_MS', cliProvider ? AGENT_CLI_LOOKUP_DEADLINE_MS : AGENT_LOOKUP_DEADLINE_MS, env);
     const researchDeadline = resolveAgentDeadlineMs('DQL_AGENT_RESEARCH_DEADLINE_MS', cliProvider ? AGENT_CLI_RESEARCH_DEADLINE_MS : AGENT_RESEARCH_DEADLINE_MS, env);
-    if (request.requestedMode === 'research' || request.analysisDepth === 'deep') {
-        return researchDeadline;
+    // The "Thinking" control sends a thinkingMode; explicit reasoningEffort/
+    // analysisDepth (CLI flags) take precedence but the mode is the common case.
+    // The deadline MUST see the resolved values — "High" both slows every call
+    // AND means deep — or picking High reliably times out at the lookup budget.
+    const resolvedMode = request.thinkingMode ? resolveThinkingMode(request.thinkingMode) : {};
+    const effectiveEffort = request.reasoningEffort ?? resolvedMode.reasoningEffort;
+    const effectiveDepth = request.analysisDepth ?? resolvedMode.analysisDepth;
+    // The budget is a CEILING, not a target — a fast answer returns immediately;
+    // the larger tiers only matter when the build genuinely needs the time.
+    const scale = (ms) => Math.min(600_000, Math.round(ms * reasoningEffortBudgetMultiplier(effectiveEffort)));
+    if (request.requestedMode === 'research' || effectiveDepth === 'deep') {
+        return scale(researchDeadline);
     }
     const plan = buildAnalysisQuestionPlan(request.question);
-    return plan.needsResearchWorkspace ? researchDeadline : lookupDeadline;
+    if (plan.needsResearchWorkspace || isAnalyticalBuildQuestion(plan)) {
+        return scale(researchDeadline);
+    }
+    return scale(lookupDeadline);
 }
 export function shouldSynthesizeAgentRunAnswer(governedAnswer) {
     if (governedAnswer.kind === 'no_answer')
@@ -344,9 +383,12 @@ export function conversationTurnInputFromRun(run) {
     // value window. Deriving dimensions from only the eight preview rows caused a
     // valid row 9/10 member to disappear before the next turn (AGT-012/E2E-010).
     const memberRows = Array.isArray(result?.rows)
-        ? result.rows.filter((row) => Boolean(row && typeof row === 'object' && !Array.isArray(row))).slice(0, 24)
+        ? result.rows.filter((row) => Boolean(row && typeof row === 'object' && !Array.isArray(row))).slice(0, 50)
         : [];
-    const rows = memberRows.slice(0, 8);
+    // The full retained window feeds both member resolution and cross-result
+    // follow-up compute ("of the results above, the average"). The store clamps
+    // it (MAX_SAMPLE_ROWS); rows are never dumped into the prompt.
+    const rows = memberRows;
     const rowsSample = rows.map((row) => columns.map((column) => row[column]));
     const contextPack = agentRunRecord(payload?.contextPack);
     const questionPlan = agentRunRecord(contextPack?.questionPlan);
@@ -8549,8 +8591,11 @@ export async function startLocalServer(opts) {
                 res.end(serializeJSON(preview));
             }
             catch (error) {
+                const raw = error instanceof Error ? error.message : String(error);
                 res.writeHead(500, { 'Content-Type': 'application/json; charset=utf-8' });
-                res.end(serializeJSON({ error: error instanceof Error ? error.message : String(error) }));
+                // `error` stays the full compiler output (the UI shows it in a
+                // collapsible); `friendlyMessage` is the one actionable sentence.
+                res.end(serializeJSON({ error: raw, friendlyMessage: compactSemanticRuntimeFailure(raw) }));
             }
             return;
         }
@@ -9095,6 +9140,26 @@ export async function startLocalServer(opts) {
             }
             return;
         }
+        // Set the runtime preference alone (no dbt Cloud test required), so a user
+        // can prefer Native / Local MetricFlow / dbt Cloud explicitly.
+        if (req.method === 'POST' && path === '/api/semantic-runtime/preference') {
+            try {
+                const body = await readJSON(req);
+                const preference = body.preference;
+                if (preference !== 'auto' && preference !== 'native' && preference !== 'metricflow-cli' && preference !== 'dbt-cloud') {
+                    throw new Error('preference must be one of: auto, native, metricflow-cli, dbt-cloud');
+                }
+                const settings = saveSemanticRuntimePreference(projectRoot, preference);
+                const runtime = await getSemanticRuntimeStatus(projectRoot, { probeConfiguredCloud: true });
+                res.writeHead(200, { 'Content-Type': 'application/json; charset=utf-8' });
+                res.end(serializeJSON({ ok: true, ...settings, runtime }));
+            }
+            catch (error) {
+                res.writeHead(400, { 'Content-Type': 'application/json; charset=utf-8' });
+                res.end(serializeJSON({ ok: false, error: error instanceof Error ? error.message : String(error) }));
+            }
+            return;
+        }
         // ── Semantic layer discovery API ─────────────────────────────────────────
         if (req.method === 'GET' && path === '/api/semantic-layer') {
             const userPrefs = readUserPrefs(userPrefsPath);
@@ -9588,7 +9653,8 @@ export async function startLocalServer(opts) {
                 .split(',')
                 .map((value) => value.trim())
                 .filter(Boolean);
-            const dimensions = (await listRuntimeCompatibleDimensions(projectRoot, semanticLayer, metrics)).map((d) => ({
+            const compat = await describeRuntimeCompatibility(projectRoot, semanticLayer, metrics, projectConfig);
+            const dimensions = compat.dimensions.map((d) => ({
                 name: d.name,
                 label: d.label,
                 description: d.description,
@@ -9598,9 +9664,19 @@ export async function startLocalServer(opts) {
                 table: d.table,
                 tags: d.tags ?? [],
                 owner: d.owner ?? null,
+                // Additive: qualified name, real grains, and time flag (existing fields
+                // above are byte-identical for compatibility).
+                qualifiedName: d.qualifiedName ?? null,
+                granularities: d.granularities ?? d.granularities ?? null,
+                isTimeDimension: Boolean(d.isTimeDimension),
             }));
             res.writeHead(200, { 'Content-Type': 'application/json; charset=utf-8' });
-            res.end(serializeJSON({ dimensions }));
+            res.end(serializeJSON({
+                dimensions,
+                engine: compat.engine,
+                incompatible: compat.incompatible,
+                degraded: compat.degraded ?? null,
+            }));
             return;
         }
         if (req.method === 'GET' && path === '/api/user-prefs/favorites') {