Make recall feedback runtime configurable from the control portal

.github/git-vibe.yml

@@ -151,6 +151,7 @@ ai:
           required: true
           tools: 
             - recall_memory
+            - submit_recall_feedback
             - assemble_context
             - get_memory
             - get_fact
@@ -165,6 +166,7 @@ ai:
           required: true
           tools: 
             - recall_memory
+            - submit_recall_feedback
             - assemble_context
             - get_memory
             - get_fact
@@ -179,6 +181,7 @@ ai:
           required: true
           tools: 
             - recall_memory
+            - submit_recall_feedback
             - assemble_context
             - get_memory
             - get_fact
@@ -193,10 +196,11 @@ ai:
           required: true
           tools: 
             - recall_memory
+            - submit_recall_feedback
             - assemble_context
             - get_memory
             - get_fact
             - keyword_search
             - semantic_search
             - list_communities
-            - get_community_summary
+            - get_community_summary

README.md

@@ -161,9 +161,9 @@ scrape job when Prometheus is shared.
 
 Online recall-quality cards use `densemem_recall_feedback_total` and
 `densemem_recall_feedback_quality_score`. They stay at zero until
-`RECALL_FEEDBACK_ENABLED=true` registers the `submit_recall_feedback` tool and a
-host LLM submits compact feedback for `recall_memory` results. Normal production
-recall traffic still contributes request volume, result count, and latency.
+recall feedback is enabled from the control portal config panel and a host LLM
+submits compact feedback for `recall_memory` results. Normal production recall
+traffic still contributes request volume, result count, and latency.
 
 For the disposable demo image, keep the control portal disabled and use the
 demo telemetry overlay instead:

README.zh-CN.md

@@ -37,7 +37,7 @@
 </p>
 
 <p align="center">
-  <a href="https://zenodo.org/records/20469578"><img src="https://zenodo.org/badge/DOI/10.5281/zenodo.20469578.svg" alt="DOI: 10.5281/zenodo.20469578" /></a>
+  <a href="https://zenodo.org/records/20519039"><img src="https://zenodo.org/badge/DOI/10.5281/zenodo.20519039.svg" alt="DOI: 10.5281/zenodo.20519039" /></a>
 </p>
 
 Dense-Mem 是一层给 MCP 客户端使用的持久记忆服务。它把来源、claims、facts、
@@ -50,7 +50,7 @@ Dense-Mem 负责把记忆状态存稳、管住，并返回可以解释给用户
 
 Dense-Mem 也是这篇研究预印本的一部分：
 [Governed Enterprise AI Memory Beyond RAG: From Vector Retrieval to Permissioned
-Knowledge Graphs](https://zenodo.org/records/20469578)。
+Knowledge Graphs](https://zenodo.org/records/20519039)。
 
 ## 项目介绍视频
 
@@ -123,6 +123,51 @@ URL、model 和 dimensions 提供 OpenAI 默认值：`https://api.openai.com/v1`
 `text-embedding-3-small`、`1536`。因此最小本地部署只需要补上 `AI_API_KEY`。
 如果切换到其他 embedding provider 或 model，请一起覆盖这些配置。
 
+### Telemetry Overlay
+
+Prometheus telemetry 默认关闭，是可选功能。要为 `/ui` 应用和 control portal
+dashboards 收集 usage、performance、verifier token、embedding token、recall
+和 promotion metrics，可以把基础 stack 和 telemetry overlay 一起启动：
+
+```bash
+curl -fsSLo prometheus.yml \
+  https://raw.githubusercontent.com/markhuangai/dense-mem/main/examples/prometheus.yml
+curl -fsSLo docker-compose.telemetry.yml \
+  https://raw.githubusercontent.com/markhuangai/dense-mem/main/examples/docker-compose.telemetry.yml
+
+export TELEMETRY_SCRAPE_TOKEN="$(openssl rand -hex 32)"
+docker compose -f docker-compose.yml -f docker-compose.telemetry.yml up -d
+```
+
+这个 overlay 会在 `127.0.0.1:9090` 启动 Prometheus，保留 30 天样本，把
+`TELEMETRY_SCRAPE_TOKEN` 作为 scrape secret 传给 Prometheus，并让 Dense-Mem
+使用 `http://prometheus:9090` 查询 telemetry。它也会设置
+`TELEMETRY_PROMETHEUS_JOB=dense-mem`，这样在共享 Prometheus 时 dashboard
+只查询 `dense-mem` scrape job。
+
+在线 recall quality 卡片使用 `densemem_recall_feedback_total` 和
+`densemem_recall_feedback_quality_score`。在 control portal config panel
+开启 recall feedback，并且宿主 LLM 为 `recall_memory` 结果提交 compact
+feedback 之前，这些卡片会保持为零。正常生产 recall 流量仍然会贡献请求量、
+结果数和延迟指标。
+
+对于一次性 demo image，保持 control portal 关闭，改用 demo telemetry
+overlay：
+
+```bash
+curl -fsSLo prometheus.demo.yml \
+  https://raw.githubusercontent.com/markhuangai/dense-mem/main/examples/prometheus.demo.yml
+curl -fsSLo docker-compose.demo.telemetry.yml \
+  https://raw.githubusercontent.com/markhuangai/dense-mem/main/examples/docker-compose.demo.telemetry.yml
+
+export TELEMETRY_SCRAPE_TOKEN="$(openssl rand -hex 32)"
+docker compose -f docker-compose.yml -f docker-compose.demo.telemetry.yml up -d
+```
+
+demo overlay 会在私有 Compose network 上 scrape `demo:8091`，并设置
+`TELEMETRY_PROMETHEUS_JOB=dense-mem-demo`。不要把这个 metrics listener
+公开到公网。
+
 ## 能力对比
 
 | 能力 | Dense-Mem | 文件记忆 | Vector DB | 通用 MCP memory |

cmd/demo-server/main.go

@@ -50,6 +50,8 @@ type scopedReaderAdapter struct {
 	inner neo4j.ScopedReader
 }
 
+const startupTimeout = 5 * time.Minute
+
 func (a *scopedReaderAdapter) ScopedRead(ctx context.Context, profileID string, query string, params map[string]any) (any, []map[string]any, error) {
 	summary, rows, err := a.inner.ScopedRead(ctx, profileID, query, params)
 	return summary, rows, err
@@ -108,9 +110,8 @@ func main() {
 	validation.SetEmbeddingDimensions(cfg.GetEmbeddingDimensions())
 	middleware.SetAuthVerificationConcurrency(cfg.AuthVerifyMaxConcurrency)
 
-	// Create root context with timeout for startup. A cold database may need
-	// time to apply migrations before schema-dependent checks can run.
-	startupCtx, startupCancel := context.WithTimeout(context.Background(), 60*time.Second)
+	// A cold Neo4j instance can need several minutes to create schema indexes.
+	startupCtx, startupCancel := context.WithTimeout(context.Background(), startupTimeout)
 	defer startupCancel()
 
 	// Initialize Postgres connection (REQUIRED for production)
@@ -488,7 +489,7 @@ func main() {
 		FragmentList:                fragmentListSvc,
 		Recall:                      recallRegistrySvc,
 		Metrics:                     discoverabilityMetrics,
-		RecallFeedbackEnabled:       cfg.GetRecallFeedbackEnabled(),
+		RecallFeedbackConfig:        appConfigService,
 		KeywordSearch:               keywordSearchService,
 		SemanticSearch:              semanticSearchService,
 		GraphQuery:                  graphQueryService,
@@ -554,10 +555,10 @@ func main() {
 	factRetractHandler := handler.NewFactRetractHandler(factRetractSvc)
 	communityReadHandler := handler.NewCommunityReadHandler(communityGetSvc)
 	communityListHandler := handler.NewCommunityListHandler(communityListSvc)
-	toolCatalogHandler := handler.NewToolCatalogHandler(toolRegistry)
-	toolReadHandler := handler.NewToolReadHandler(toolRegistry)
-	toolExecuteHandler := handler.NewToolExecuteHandler(toolRegistry)
-	mcpHandler := handler.NewMCPHandlerWithLifecycle(toolRegistry, logger, streamLifecycle)
+	toolCatalogHandler := handler.NewToolCatalogHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	toolReadHandler := handler.NewToolReadHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	toolExecuteHandler := handler.NewToolExecuteHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	mcpHandler := handler.NewMCPHandlerWithLifecycleAndRuntimeConfig(toolRegistry, logger, streamLifecycle, appConfigService)
 	openAPIAISafeHandler := handler.NewOpenAPIHandler(openAPIGen, openapi.SpecVariantAISafe)
 	openAPIFullHandler := handler.NewOpenAPIHandler(openAPIGen, openapi.SpecVariantFull)
 

cmd/migrate/main.go

@@ -14,6 +14,8 @@ import (
 	"github.com/markhuangai/dense-mem/internal/storage/postgres"
 )
 
+const migrationTimeout = 5 * time.Minute
+
 func main() {
 	if len(os.Args) < 2 {
 		fmt.Fprintf(os.Stderr, "Usage: %s <up|down|status|backfill-neo4j>\n", os.Args[0])
@@ -36,8 +38,7 @@ func main() {
 		log.Fatalf("Failed to load config: %v", err)
 	}
 
-	// Create context with timeout
-	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), migrationTimeout)
 	defer cancel()
 
 	// Execute command

cmd/server/main.go

@@ -48,6 +48,8 @@ type scopedReaderAdapter struct {
 	inner neo4j.ScopedReader
 }
 
+const startupTimeout = 5 * time.Minute
+
 func (a *scopedReaderAdapter) ScopedRead(ctx context.Context, profileID string, query string, params map[string]any) (any, []map[string]any, error) {
 	summary, rows, err := a.inner.ScopedRead(ctx, profileID, query, params)
 	return summary, rows, err
@@ -101,9 +103,8 @@ func main() {
 	validation.SetEmbeddingDimensions(cfg.GetEmbeddingDimensions())
 	middleware.SetAuthVerificationConcurrency(cfg.AuthVerifyMaxConcurrency)
 
-	// Create root context with timeout for startup. A cold database may need
-	// time to apply migrations before schema-dependent checks can run.
-	startupCtx, startupCancel := context.WithTimeout(context.Background(), 60*time.Second)
+	// A cold Neo4j instance can need several minutes to create schema indexes.
+	startupCtx, startupCancel := context.WithTimeout(context.Background(), startupTimeout)
 	defer startupCancel()
 
 	// Initialize Postgres connection (REQUIRED for production)
@@ -436,7 +437,7 @@ func main() {
 		FragmentList:                fragmentListSvc,
 		Recall:                      recallRegistrySvc,
 		Metrics:                     discoverabilityMetrics,
-		RecallFeedbackEnabled:       cfg.GetRecallFeedbackEnabled(),
+		RecallFeedbackConfig:        appConfigService,
 		KeywordSearch:               keywordSearchService,
 		SemanticSearch:              semanticSearchService,
 		GraphQuery:                  graphQueryService,
@@ -502,10 +503,10 @@ func main() {
 	factRetractHandler := handler.NewFactRetractHandler(factRetractSvc)
 	communityReadHandler := handler.NewCommunityReadHandler(communityGetSvc)
 	communityListHandler := handler.NewCommunityListHandler(communityListSvc)
-	toolCatalogHandler := handler.NewToolCatalogHandler(toolRegistry)
-	toolReadHandler := handler.NewToolReadHandler(toolRegistry)
-	toolExecuteHandler := handler.NewToolExecuteHandler(toolRegistry)
-	mcpHandler := handler.NewMCPHandlerWithLifecycle(toolRegistry, logger, streamLifecycle)
+	toolCatalogHandler := handler.NewToolCatalogHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	toolReadHandler := handler.NewToolReadHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	toolExecuteHandler := handler.NewToolExecuteHandlerWithRuntimeConfig(toolRegistry, appConfigService)
+	mcpHandler := handler.NewMCPHandlerWithLifecycleAndRuntimeConfig(toolRegistry, logger, streamLifecycle, appConfigService)
 	openAPIAISafeHandler := handler.NewOpenAPIHandler(openAPIGen, openapi.SpecVariantAISafe)
 	openAPIFullHandler := handler.NewOpenAPIHandler(openAPIGen, openapi.SpecVariantFull)
 

internal/config/config.go

@@ -89,7 +89,6 @@ type Config struct {
 	ClaimWriteRateLimit          int
 	ClaimReadRateLimit           int
 	RecallValidatedClaimWeight   float64
-	RecallFeedbackEnabled        bool
 	PromoteTxTimeoutSeconds      int
 	SkillPackImportHistoryDays   int
 	AICommunityMaxNodes          int
@@ -155,7 +154,6 @@ func (c *Config) GetAIVerifierMaxConcurrency() int       { return c.AIVerifierMa
 func (c *Config) GetClaimWriteRateLimit() int            { return c.ClaimWriteRateLimit }
 func (c *Config) GetClaimReadRateLimit() int             { return c.ClaimReadRateLimit }
 func (c *Config) GetRecallValidatedClaimWeight() float64 { return c.RecallValidatedClaimWeight }
-func (c *Config) GetRecallFeedbackEnabled() bool         { return c.RecallFeedbackEnabled }
 func (c *Config) GetPromoteTxTimeoutSeconds() int        { return c.PromoteTxTimeoutSeconds }
 func (c *Config) GetSkillPackImportHistoryDays() int     { return c.SkillPackImportHistoryDays }
 func (c *Config) GetAICommunityMaxNodes() int            { return c.AICommunityMaxNodes }
@@ -400,11 +398,6 @@ func Load() (Config, error) {
 		return cfg, err
 	}
 
-	cfg.RecallFeedbackEnabled, err = parseBoolOrDefault("RECALL_FEEDBACK_ENABLED", false)
-	if err != nil {
-		return cfg, err
-	}
-
 	cfg.PromoteTxTimeoutSeconds, err = parseIntOrDefault("PROMOTE_TX_TIMEOUT_SECONDS", 10)
 	if err != nil {
 		return cfg, err

internal/config/config_test.go

@@ -40,7 +40,6 @@ func clearEnv() {
 		"CLAIM_WRITE_RATE_LIMIT",
 		"CLAIM_READ_RATE_LIMIT",
 		"RECALL_VALIDATED_CLAIM_WEIGHT",
-		"RECALL_FEEDBACK_ENABLED",
 		"PROMOTE_TX_TIMEOUT_SECONDS",
 		"AI_COMMUNITY_MAX_NODES",
 		"CONTROL_HTTP_ADDR",
@@ -134,9 +133,6 @@ func TestLoadDefaults(t *testing.T) {
 	if cfg.TelemetryPrometheusJob != "" {
 		t.Errorf("TelemetryPrometheusJob default = %q, want empty", cfg.TelemetryPrometheusJob)
 	}
-	if cfg.GetRecallFeedbackEnabled() {
-		t.Error("GetRecallFeedbackEnabled() default = true, want false")
-	}
 }
 
 func TestLoadTelemetryConfig(t *testing.T) {
@@ -193,21 +189,6 @@ func TestLoadTelemetryConfigRequiresScrapeToken(t *testing.T) {
 	}
 }
 
-func TestLoadRecallFeedbackConfig(t *testing.T) {
-	clearEnv()
-	setRequiredEnv()
-	os.Setenv("RECALL_FEEDBACK_ENABLED", "true")
-
-	cfg, err := Load()
-	if err != nil {
-		t.Fatalf("Load() returned unexpected error: %v", err)
-	}
-
-	if !cfg.GetRecallFeedbackEnabled() {
-		t.Fatal("GetRecallFeedbackEnabled() = false, want true")
-	}
-}
-
 func TestLoadValidation_MissingPostgresDSN(t *testing.T) {
 	clearEnv()
 	os.Setenv("NEO4J_URI", "bolt://localhost:7687")

internal/domain/app_config.go

@@ -28,6 +28,8 @@ const (
 	AppConfigCommunityDetectionJitterSeconds  = "COMMUNITY_DETECTION_JITTER_SECONDS"
 
 	AppConfigOperationLogRetentionDays = "OPERATION_LOG_RETENTION_DAYS"
+
+	AppConfigRecallFeedbackEnabled = "RECALL_FEEDBACK_ENABLED"
 )
 
 type AppConfigEntry struct {
@@ -89,3 +91,24 @@ type OperationLogConfigItem struct {
 type OperationLogRuntimeConfig struct {
 	RetentionDays int `json:"retention_days"`
 }
+
+// RecallFeedbackConfigSettings is the editable runtime configuration for
+// host-LLM recall feedback telemetry.
+type RecallFeedbackConfigSettings struct {
+	UpdateTime string                      `json:"update_time"`
+	Items      []RecallFeedbackConfigItem  `json:"items"`
+	Effective  RecallFeedbackRuntimeConfig `json:"effective"`
+}
+
+// RecallFeedbackConfigItem is one control-panel editable recall feedback config entry.
+type RecallFeedbackConfigItem struct {
+	Key            string    `json:"key"`
+	Value          string    `json:"value"`
+	EffectiveValue string    `json:"effective_value"`
+	UpdatedAt      time.Time `json:"updated_at"`
+}
+
+// RecallFeedbackRuntimeConfig is the effective recall feedback runtime config.
+type RecallFeedbackRuntimeConfig struct {
+	Enabled bool `json:"enabled"`
+}

internal/http/control_portal.go

@@ -181,6 +181,8 @@ func NewControlPortalServerWithMetricsAndTelemetry(
 		api.PATCH("/config/community-detection", control.updateCommunityDetectionConfig)
 		api.GET("/config/operation-logs", control.getOperationLogConfig)
 		api.PATCH("/config/operation-logs", control.updateOperationLogConfig)
+		api.GET("/config/recall-feedback", control.getRecallFeedbackConfig)
+		api.PATCH("/config/recall-feedback", control.updateRecallFeedbackConfig)
 	}
 	if telemetry.SSO != nil {
 		api.GET("/sso/providers", control.listSSOProviders)