Thread per-tenant AI narrators into pulse_summarize via chat session

v1 of pulse_summarize (1fe5d6853) shipped with heuristic narrative
only. The follow-up wiring promised in that commit now lands: the
chat session carries optional report-narration providers that the
tool's handler reads when building requests, so AI-narrated synthesis
flows into chat using the same provider, sanitizer, model selection,
cost ledger, and budget gate the report PDF endpoint already uses.

Pipeline:
- pkg/reporting Narrator / FleetNarrator / FindingsProvider interfaces
  are already implemented by internal/ai.Service. No new
  implementations.
- tools.ExecutorConfig + PulseToolExecutor gain three optional fields
  (ReportNarrator, ReportFleetNarrator, ReportFindingsProvider).
  Clone() copies them so per-session executors inherit the wiring.
- chat.Config gains the same three fields; NewService threads them
  into ExecutorConfig.
- tools_summarize.go reads e.reportNarrator/FleetNarrator/
  FindingsProvider and populates MetricReportRequest /
  MultiReportRequest. The engine already accepts these on the request
  and falls back to heuristic when they are nil — no engine changes
  needed.
- AIHandler gains SetReportNarratorResolver(ctx -> narrators); both
  per-tenant and default chat.Config construction sites invoke the
  resolver. Router wires the resolver to AISettingsHandler.GetAIService
  with the same Enabled-gate the reporting handler uses.

Unconfigured tenants are unchanged: the resolver returns nil, the
tool returns heuristic narrative — identical to today. Configured
tenants get AI synthesis in chat that matches what their report PDF
already carries, billed and budget-gated the same way.

internal/ai/chat/service.go

@@ -19,6 +19,7 @@ import (
 	"github.com/rcourtman/pulse-go-rewrite/internal/config"
 	"github.com/rcourtman/pulse-go-rewrite/internal/models"
 	"github.com/rcourtman/pulse-go-rewrite/internal/unifiedresources"
+	"github.com/rcourtman/pulse-go-rewrite/pkg/reporting"
 	"github.com/rs/zerolog/log"
 )
 
@@ -78,6 +79,14 @@ type Config struct {
 	// Stored config may say autonomous, but runtime execution must use the
 	// entitlement-clamped level.
 	ControlLevelResolver func(*config.AIConfig) string
+
+	// Optional report-narration providers for the pulse_summarize tool.
+	// When the per-tenant AI service is configured the API layer passes
+	// it here for all three roles; when unconfigured the tool returns
+	// heuristic narrative.
+	ReportNarrator         reporting.Narrator
+	ReportFleetNarrator    reporting.FleetNarrator
+	ReportFindingsProvider reporting.FindingsProvider
 }
 
 // Service provides direct AI chat without external sidecar
@@ -138,6 +147,9 @@ func NewService(cfg Config) *Service {
 		AgentServer:            agentServer,
 		RecoveryPointsProvider: cfg.RecoveryPointsProvider,
 		OrgID:                  cfg.OrgID,
+		ReportNarrator:         cfg.ReportNarrator,
+		ReportFleetNarrator:    cfg.ReportFleetNarrator,
+		ReportFindingsProvider: cfg.ReportFindingsProvider,
 	}
 
 	if cfg.AIConfig != nil {

internal/ai/tools/executor.go

@@ -10,6 +10,7 @@ import (
 	"github.com/rcourtman/pulse-go-rewrite/internal/models"
 	"github.com/rcourtman/pulse-go-rewrite/internal/recovery"
 	"github.com/rcourtman/pulse-go-rewrite/internal/unifiedresources"
+	"github.com/rcourtman/pulse-go-rewrite/pkg/reporting"
 	"github.com/rs/zerolog/log"
 )
 
@@ -486,6 +487,19 @@ type ExecutorConfig struct {
 	ControlLevel    ControlLevel
 	ProtectedGuests []string // VMIDs that AI cannot control
 	OrgID           string   // Tenant/org scope for approval records
+
+	// Optional report-narration providers, used by the pulse_summarize
+	// tool to produce AI-narrated synthesis in chat. When the per-tenant
+	// AI service is configured these are the same interfaces it exposes
+	// to the reporting handler for PDF generation. Absent values cause
+	// the tool to return heuristic narrative instead — identical to
+	// what the report PDF carries when no AI is configured. Defined
+	// here (rather than as separate setters) so chat.Config can wire
+	// them at session construction time alongside the rest of the
+	// executor's providers.
+	ReportNarrator         reporting.Narrator
+	ReportFleetNarrator    reporting.FleetNarrator
+	ReportFindingsProvider reporting.FindingsProvider
 }
 
 // PulseToolExecutor implements ToolExecutor for Pulse-specific tools
@@ -566,6 +580,14 @@ type PulseToolExecutor struct {
 	patrolFindingCreatorMu sync.RWMutex
 	patrolFindingCreator   PatrolFindingCreator
 
+	// Report-narration providers, used by pulse_summarize when the
+	// per-tenant AI service is configured. Absent values cause the tool
+	// to fall back to heuristic narrative — identical to the report PDF
+	// behaviour when AI is unconfigured.
+	reportNarrator         reporting.Narrator
+	reportFleetNarrator    reporting.FleetNarrator
+	reportFindingsProvider reporting.FindingsProvider
+
 	// Tool registry
 	registry *ToolRegistry
 }
@@ -621,6 +643,9 @@ func NewPulseToolExecutor(cfg ExecutorConfig) *PulseToolExecutor {
 		controlLevel:               cfg.ControlLevel,
 		protectedGuests:            cfg.ProtectedGuests,
 		orgID:                      normalizeExecutorOrgID(cfg.OrgID),
+		reportNarrator:             cfg.ReportNarrator,
+		reportFleetNarrator:        cfg.ReportFleetNarrator,
+		reportFindingsProvider:     cfg.ReportFindingsProvider,
 		registry:                   NewToolRegistry(),
 	}
 
@@ -698,6 +723,9 @@ func (e *PulseToolExecutor) Clone() *PulseToolExecutor {
 		isAutonomous:               e.isAutonomous,
 		orgID:                      e.orgID,
 		telemetryCallback:          e.telemetryCallback,
+		reportNarrator:             e.reportNarrator,
+		reportFleetNarrator:        e.reportFleetNarrator,
+		reportFindingsProvider:     e.reportFindingsProvider,
 		registry:                   e.registry,
 	}
 	clone.patrolFindingCreator = e.GetPatrolFindingCreator()

internal/ai/tools/tools_summarize.go

@@ -155,10 +155,12 @@ func (e *PulseToolExecutor) summarizeResource(
 	}
 
 	req := reporting.MetricReportRequest{
-		ResourceType: canonicalType,
-		ResourceID:   resourceID,
-		Start:        start,
-		End:          end,
+		ResourceType:     canonicalType,
+		ResourceID:       resourceID,
+		Start:            start,
+		End:              end,
+		Narrator:         e.reportNarrator,
+		FindingsProvider: e.reportFindingsProvider,
 	}
 	narrative, err := engine.NarrativeFor(req)
 	if err != nil {
@@ -255,10 +257,13 @@ func (e *PulseToolExecutor) summarizeFleet(
 	}
 
 	req := reporting.MultiReportRequest{
-		Title:     "Fleet summary",
-		Start:     start,
-		End:       end,
-		Resources: resources,
+		Title:            "Fleet summary",
+		Start:            start,
+		End:              end,
+		Resources:        resources,
+		FleetNarrator:    e.reportFleetNarrator,
+		Narrator:         e.reportNarrator,
+		FindingsProvider: e.reportFindingsProvider,
 	}
 	narrative, err := engine.FleetNarrativeFor(req)
 	if err != nil {

internal/ai/tools/tools_summarize_test.go

@@ -230,6 +230,158 @@ func TestSummarizeTool_FleetEnforcesMaxResources(t *testing.T) {
 	}
 }
 
+// stubReportNarrator implements reporting.Narrator with a recorded
+// invocation so the test can assert the tool delegates to it.
+type stubReportNarrator struct {
+	called   bool
+	seen     reporting.NarrativeInput
+	response reporting.Narrative
+}
+
+func (s *stubReportNarrator) Narrate(_ context.Context, in reporting.NarrativeInput) (reporting.Narrative, error) {
+	s.called = true
+	s.seen = in
+	return s.response, nil
+}
+
+type stubFleetReportNarrator struct {
+	called   bool
+	seen     reporting.FleetNarrativeInput
+	response reporting.FleetNarrative
+}
+
+func (s *stubFleetReportNarrator) NarrateFleet(_ context.Context, in reporting.FleetNarrativeInput) (reporting.FleetNarrative, error) {
+	s.called = true
+	s.seen = in
+	return s.response, nil
+}
+
+func TestSummarizeTool_UsesReportNarratorWhenConfigured(t *testing.T) {
+	dir := t.TempDir()
+	store, err := metrics.NewStore(metrics.StoreConfig{
+		DBPath:          filepath.Join(dir, "metrics.db"),
+		WriteBufferSize: 10,
+		FlushInterval:   50 * time.Millisecond,
+		RetentionRaw:    24 * time.Hour,
+		RetentionMinute: 7 * 24 * time.Hour,
+		RetentionHourly: 30 * 24 * time.Hour,
+		RetentionDaily:  90 * 24 * time.Hour,
+	})
+	if err != nil {
+		t.Fatalf("metrics store: %v", err)
+	}
+	defer store.Close()
+	engine := reporting.NewReportEngine(reporting.EngineConfig{MetricsStore: store})
+	prev := reporting.GetEngine()
+	reporting.SetEngine(engine)
+	defer reporting.SetEngine(prev)
+
+	narrator := &stubReportNarrator{
+		response: reporting.Narrative{
+			Source:        reporting.NarrativeSourceAI,
+			HealthStatus:  "HEALTHY",
+			HealthMessage: "AI says fine",
+			Observations: []reporting.NarrativeBullet{
+				{Text: "AI bullet", Severity: reporting.NarrativeSeverityOK},
+			},
+			Recommendations: []string{"Continue monitoring"},
+		},
+	}
+	exec := NewPulseToolExecutor(ExecutorConfig{
+		ReportNarrator: narrator,
+	})
+
+	res, err := exec.executeSummarize(context.Background(), map[string]interface{}{
+		"action":        "resource",
+		"resource_type": "node",
+		"resource_id":   "node-with-narrator",
+	})
+	if err != nil {
+		t.Fatalf("executeSummarize: %v", err)
+	}
+	if res.IsError {
+		t.Fatalf("unexpected error: %+v", res.Content)
+	}
+	if !narrator.called {
+		t.Fatal("expected narrator to be invoked")
+	}
+	var parsed summarizeResourceResponse
+	if err := json.Unmarshal([]byte(res.Content[0].Text), &parsed); err != nil {
+		t.Fatalf("decode response: %v", err)
+	}
+	if parsed.NarrativeSource != reporting.NarrativeSourceAI {
+		t.Errorf("NarrativeSource = %q, want ai", parsed.NarrativeSource)
+	}
+	if parsed.HealthMessage != "AI says fine" {
+		t.Errorf("HealthMessage = %q, want AI says fine", parsed.HealthMessage)
+	}
+	if len(parsed.Observations) != 1 || parsed.Observations[0].Text != "AI bullet" {
+		t.Errorf("Observations = %#v", parsed.Observations)
+	}
+}
+
+func TestSummarizeTool_FleetUsesFleetNarratorWhenConfigured(t *testing.T) {
+	dir := t.TempDir()
+	store, err := metrics.NewStore(metrics.StoreConfig{
+		DBPath:          filepath.Join(dir, "metrics.db"),
+		WriteBufferSize: 10,
+		FlushInterval:   50 * time.Millisecond,
+		RetentionRaw:    24 * time.Hour,
+		RetentionMinute: 7 * 24 * time.Hour,
+		RetentionHourly: 30 * 24 * time.Hour,
+		RetentionDaily:  90 * 24 * time.Hour,
+	})
+	if err != nil {
+		t.Fatalf("metrics store: %v", err)
+	}
+	defer store.Close()
+	engine := reporting.NewReportEngine(reporting.EngineConfig{MetricsStore: store})
+	prev := reporting.GetEngine()
+	reporting.SetEngine(engine)
+	defer reporting.SetEngine(prev)
+
+	fleet := &stubFleetReportNarrator{
+		response: reporting.FleetNarrative{
+			Source:           reporting.NarrativeSourceAI,
+			HealthStatus:     "WARNING",
+			HealthMessage:    "Memory creeping up",
+			ExecutiveSummary: "AI fleet summary text",
+			Outliers: []reporting.FleetOutlier{
+				{ResourceID: "node-a", ResourceName: "alpha", Reason: "Memory at 92%", Severity: reporting.NarrativeSeverityWarning},
+			},
+			Recommendations: []string{"Review memory allocation"},
+		},
+	}
+	exec := NewPulseToolExecutor(ExecutorConfig{
+		ReportFleetNarrator: fleet,
+	})
+
+	res, err := exec.executeSummarize(context.Background(), map[string]interface{}{
+		"action":        "fleet",
+		"resource_type": "node",
+		"resource_ids":  "node-a,node-b",
+	})
+	if err != nil {
+		t.Fatalf("executeSummarize: %v", err)
+	}
+	if res.IsError {
+		t.Fatalf("unexpected error: %+v", res.Content)
+	}
+	if !fleet.called {
+		t.Fatal("expected fleet narrator to be invoked")
+	}
+	var parsed summarizeFleetResponse
+	if err := json.Unmarshal([]byte(res.Content[0].Text), &parsed); err != nil {
+		t.Fatalf("decode response: %v", err)
+	}
+	if parsed.NarrativeSource != reporting.NarrativeSourceAI {
+		t.Errorf("NarrativeSource = %q, want ai", parsed.NarrativeSource)
+	}
+	if len(parsed.Outliers) != 1 || parsed.Outliers[0].ResourceName != "alpha" {
+		t.Errorf("Outliers = %#v", parsed.Outliers)
+	}
+}
+
 func TestSummarizeRangeWindow(t *testing.T) {
 	cases := map[string]time.Duration{
 		"24h":     24 * time.Hour,

internal/api/ai_handler.go

@@ -23,6 +23,7 @@ import (
 	recoverymanager "github.com/rcourtman/pulse-go-rewrite/internal/recovery/manager"
 	"github.com/rcourtman/pulse-go-rewrite/internal/unifiedresources"
 	"github.com/rcourtman/pulse-go-rewrite/pkg/aicontracts"
+	"github.com/rcourtman/pulse-go-rewrite/pkg/reporting"
 	"github.com/rs/zerolog/log"
 )
 
@@ -111,6 +112,36 @@ type AIHandler struct {
 	approvalCreatedCallback func(*approval.ApprovalRequest)
 	controlLevelResolver    func(context.Context, *config.AIConfig) string
 	patrolRunProvider       patrolRunHandoffProvider
+
+	// reportNarratorResolver returns the per-tenant report-narration
+	// interfaces (single-resource, fleet, findings) for chat sessions
+	// so the pulse_summarize tool can produce AI-narrated synthesis
+	// using the same provider, sanitizer, budget gate, and cost ledger
+	// the report PDF endpoint already uses. Wired by the router from
+	// AISettingsHandler.GetAIService; absent values cause the tool to
+	// return heuristic narrative. The ctx must carry the tenant org
+	// via GetOrgID — same convention the reporting handler uses.
+	reportNarratorResolver func(ctx context.Context) (reporting.Narrator, reporting.FleetNarrator, reporting.FindingsProvider)
+}
+
+// SetReportNarratorResolver wires the optional per-tenant
+// report-narrator resolver. When unset (or when the resolver returns
+// nil), chat sessions construct their tool executor without report
+// narrators and pulse_summarize falls back to heuristic narrative.
+func (h *AIHandler) SetReportNarratorResolver(
+	resolver func(ctx context.Context) (reporting.Narrator, reporting.FleetNarrator, reporting.FindingsProvider),
+) {
+	if h == nil {
+		return
+	}
+	h.reportNarratorResolver = resolver
+}
+
+func (h *AIHandler) resolveReportNarrator(ctx context.Context) (reporting.Narrator, reporting.FleetNarrator, reporting.FindingsProvider) {
+	if h == nil || h.reportNarratorResolver == nil {
+		return nil, nil, nil
+	}
+	return h.reportNarratorResolver(ctx)
 }
 
 // newChatService is the factory function for creating the AI service.
@@ -474,6 +505,8 @@ func (h *AIHandler) initTenantService(ctx context.Context, orgID string) AIServi
 		}
 	}
 
+	chatCfg.ReportNarrator, chatCfg.ReportFleetNarrator, chatCfg.ReportFindingsProvider = h.resolveReportNarrator(tenantCtx)
+
 	svc := newChatService(chatCfg)
 	if err := svc.Start(ctx); err != nil {
 		log.Error().Str("orgID", orgID).Err(err).Msg("Failed to start AI service for tenant")
@@ -708,6 +741,8 @@ func (h *AIHandler) Start(ctx context.Context, monitor *monitoring.Monitor) erro
 		chatCfg.RecoveryPointsProvider = tools.NewRecoveryPointsMCPAdapter(recoveryManager, orgID)
 	}
 
+	chatCfg.ReportNarrator, chatCfg.ReportFleetNarrator, chatCfg.ReportFindingsProvider = h.resolveReportNarrator(serviceCtx)
+
 	svc := newChatService(chatCfg)
 	if err := svc.Start(ctx); err != nil {
 		return fmt.Errorf("start AI chat service: %w", err)

internal/api/router.go

@@ -646,6 +646,26 @@ func (r *Router) setupRoutes() {
 	r.aiHandler.SetServiceInitializer(func(ctx context.Context, service AIService) {
 		r.wireAIChatDependenciesForService(ctx, service)
 	})
+	// Wire the per-tenant report-narrator resolver into the chat
+	// handler so the pulse_summarize tool can produce AI-narrated
+	// synthesis using the same Service that powers the report PDF
+	// endpoint. The AI service implements all three interfaces; an
+	// unconfigured tenant returns nil and the tool falls back to
+	// heuristic narrative.
+	r.aiHandler.SetReportNarratorResolver(func(ctx context.Context) (reporting.Narrator, reporting.FleetNarrator, reporting.FindingsProvider) {
+		if r.aiSettingsHandler == nil {
+			return nil, nil, nil
+		}
+		svc := r.aiSettingsHandler.GetAIService(ctx)
+		if svc == nil {
+			return nil, nil, nil
+		}
+		cfg := svc.GetAIConfig()
+		if cfg == nil || !cfg.Enabled {
+			return nil, nil, nil
+		}
+		return svc, svc, svc
+	})
 
 	// AI-powered infrastructure discovery handlers
 	// Note: The actual service is wired up later via SetDiscoveryService