feat: add webhook thought ingestion

Merge pull request 'Expose retrieval mode in query responses' (#37 ) from issue-14-expose-retrieval-mode into main
Reviewed-on: #37 Reviewed-by: Warky <2+warkanum@noreply@warky.dev>
2026-07-15 04:24:23 +02:00 · 2026-07-14 14:44:14 +00:00 · 2026-07-14 12:44:53 +02:00
14 changed files with 501 additions and 41 deletions
@@ -74,6 +74,38 @@ The AMCS directory is used to store configuration and code for the Avalon Memory
 | `describe_tools` | List all available MCP tools with names, descriptions, categories, and model-authored usage notes; call this at the start of a session to orient yourself |
 | `annotate_tool` | Persist your own usage notes for a specific tool; notes are returned by `describe_tools` in future sessions |

+## Webhook ingestion
+
+External automation can create thoughts without speaking MCP by posting JSON to `POST /webhooks/thoughts`. The endpoint is protected by the same AMCS authentication middleware as MCP and file uploads, so pass one configured API key via `x-brain-key`, an authorization bearer token header, or another enabled auth method.
+
+Example:
+
+```bash
+curl -X POST http://localhost:8080/webhooks/thoughts \
+  -H 'Content-Type: application/json' \
+  -H 'x-brain-key: <api-key>' \
+  -H 'Idempotency-Key: n8n-run-123' \
+  -d '{
+    "content": "External system observed build failure on main",
+    "project": "amcs",
+    "source": "n8n",
+    "type": "task",
+    "topics": ["ci", "webhook"],
+    "metadata": {"workflow": "ci-monitor", "run_id": "123"}
+  }'
+```
+
+Payload fields:
+
+- `content` is required and becomes the thought content.
+- `project` is optional; when present it must match an existing AMCS project.
+- `source`, `type`, `topics`, `people`, `action_items`, and `dates_mentioned` are normalized into the standard thought metadata schema. Unknown `type` values fall back to `observation`.
+- `metadata` or `source_metadata` may contain safe source-specific JSON values; unsupported values and overly deep objects are dropped rather than persisted.
+- `idempotency_key` or the `Idempotency-Key` header can be supplied to make repeated webhook deliveries return the existing thought with `duplicate: true`.
+- `external_id` is stored under `metadata.webhook.external_id` for source-side traceability.
+
+Successful new ingestion returns `201` with the created thought. Duplicate idempotency-key delivery returns `200` and the previously created thought. Invalid JSON, missing content, missing/unknown projects, or unauthenticated requests are rejected before persistence. Metadata and embedding enrichment are queued after the thought is stored.
+
 ## Learnings

 Learnings are curated, structured memory records for durable insights you want to keep distinct from raw thoughts. Use them for normalized lessons, decisions, and evidence-backed findings that should be easy to retrieve and review over time.
@@ -478,7 +510,7 @@ metadata_retry:
  include_archived: false
 ```

-**Search fallback**: when no embeddings exist for the active model in scope, `search_thoughts`, `recall_context`, `get_project_context`, `summarize_thoughts`, and `related_thoughts` automatically fall back to Postgres full-text search so results are never silently empty.
+**Search fallback**: when no embeddings exist for the active model in scope, `search_thoughts`, `recall_context`, `get_project_context`, `summarize_thoughts`, and `related_thoughts` automatically fall back to Postgres full-text search so results are never silently empty. All five tools include a `retrieval_mode` field in their response (`"semantic"` or `"text"`) so callers can see which path was taken.

 ## Client Setup

@@ -238,6 +238,7 @@ func routes(logger *slog.Logger, cfg *config.Config, info buildinfo.Info, db *st
 	}
 	mux.Handle("/files", authMiddleware(fileHandler(filesTool)))
 	mux.Handle("/files/{id}", authMiddleware(fileHandler(filesTool)))
+	mux.Handle("/webhooks/thoughts", authMiddleware(newWebhookThoughtHandler(db, embeddings, cfg.Capture, enrichmentRetryer, backfillTool)))
 	mux.HandleFunc("/.well-known/oauth-authorization-server", oauthMetadataHandler())
 	mux.HandleFunc("/api/oauth/register", oauthRegisterHandler(dynClients, logger))
 	mux.HandleFunc("/api/oauth/authorize", oauthAuthorizeHandler(dynClients, authCodes, logger))
@@ -0,0 +1,214 @@
+package app
+
+import (
+	"encoding/json"
+	"errors"
+	"net/http"
+	"strings"
+	"time"
+
+	"git.warky.dev/wdevs/amcs/internal/ai"
+	"git.warky.dev/wdevs/amcs/internal/config"
+	"git.warky.dev/wdevs/amcs/internal/metadata"
+	"git.warky.dev/wdevs/amcs/internal/store"
+	"git.warky.dev/wdevs/amcs/internal/tools"
+	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
+)
+
+const maxWebhookBodyBytes = 1 << 20
+
+type webhookThoughtRequest struct {
+	Content        string         `json:"content"`
+	Project        string         `json:"project,omitempty"`
+	Source         string         `json:"source,omitempty"`
+	Type           string         `json:"type,omitempty"`
+	Topics         []string       `json:"topics,omitempty"`
+	People         []string       `json:"people,omitempty"`
+	ActionItems    []string       `json:"action_items,omitempty"`
+	DatesMentioned []string       `json:"dates_mentioned,omitempty"`
+	Metadata       map[string]any `json:"metadata,omitempty"`
+	SourceMetadata map[string]any `json:"source_metadata,omitempty"`
+	IDempotencyKey string         `json:"idempotency_key,omitempty"`
+	ExternalID     string         `json:"external_id,omitempty"`
+}
+
+type webhookThoughtResponse struct {
+	Thought     thoughttypes.Thought         `json:"thought"`
+	Duplicate   bool                         `json:"duplicate"`
+	WebhookMeta thoughttypes.WebhookMetadata `json:"webhook"`
+}
+
+type webhookThoughtHandler struct {
+	store        *store.DB
+	embeddings   *ai.EmbeddingRunner
+	capture      config.CaptureConfig
+	retryer      tools.MetadataQueuer
+	embedRetryer tools.EmbeddingQueuer
+}
+
+func newWebhookThoughtHandler(db *store.DB, embeddings *ai.EmbeddingRunner, capture config.CaptureConfig, retryer tools.MetadataQueuer, embedRetryer tools.EmbeddingQueuer) http.Handler {
+	return &webhookThoughtHandler{store: db, embeddings: embeddings, capture: capture, retryer: retryer, embedRetryer: embedRetryer}
+}
+
+func (h *webhookThoughtHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	if r.URL.Path != "/webhooks/thoughts" {
+		http.NotFound(w, r)
+		return
+	}
+	if r.Method != http.MethodPost {
+		w.Header().Set("Allow", http.MethodPost)
+		http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	r.Body = http.MaxBytesReader(w, r.Body, maxWebhookBodyBytes)
+	in, err := parseWebhookThoughtRequest(r)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusBadRequest)
+		return
+	}
+
+	webhookMeta := buildWebhookMetadata(in, r.Header.Get("Idempotency-Key"), time.Now().UTC())
+	if webhookMeta.IDempotencyKey != "" {
+		if existing, err := h.store.GetThoughtByWebhookIDempotencyKey(r.Context(), webhookMeta.IDempotencyKey); err == nil {
+			writeWebhookThoughtResponse(w, http.StatusOK, webhookThoughtResponse{Thought: existing, Duplicate: true, WebhookMeta: webhookMeta})
+			return
+		}
+	}
+
+	projectID, err := h.resolveWebhookProject(r, in.Project)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusBadRequest)
+		return
+	}
+
+	thought := thoughttypes.Thought{
+		Content:   strings.TrimSpace(in.Content),
+		Metadata:  normalizeWebhookThoughtMetadata(in, webhookMeta, h.capture),
+		ProjectID: projectID,
+	}
+	created, err := h.store.InsertThought(r.Context(), thought, h.embeddings.PrimaryModel())
+	if err != nil {
+		http.Error(w, "insert thought: "+err.Error(), http.StatusInternalServerError)
+		return
+	}
+	if projectID != nil {
+		_ = h.store.TouchProject(r.Context(), *projectID)
+	}
+	if h.retryer != nil {
+		h.retryer.QueueThought(created.ID)
+	}
+	if h.embedRetryer != nil {
+		h.embedRetryer.QueueThought(r.Context(), created.ID, created.Content)
+	}
+
+	writeWebhookThoughtResponse(w, http.StatusCreated, webhookThoughtResponse{Thought: created, WebhookMeta: webhookMeta})
+}
+
+func parseWebhookThoughtRequest(r *http.Request) (webhookThoughtRequest, error) {
+	if !strings.Contains(r.Header.Get("Content-Type"), "application/json") {
+		return webhookThoughtRequest{}, errors.New("webhook requires application/json")
+	}
+	defer r.Body.Close()
+	decoder := json.NewDecoder(r.Body)
+	decoder.DisallowUnknownFields()
+	var in webhookThoughtRequest
+	if err := decoder.Decode(&in); err != nil {
+		return webhookThoughtRequest{}, err
+	}
+	if strings.TrimSpace(in.Content) == "" {
+		return webhookThoughtRequest{}, errors.New("content is required")
+	}
+	return in, nil
+}
+
+func (h *webhookThoughtHandler) resolveWebhookProject(r *http.Request, projectName string) (*int64, error) {
+	projectName = strings.TrimSpace(projectName)
+	if projectName == "" {
+		return nil, nil
+	}
+	project, err := h.store.GetProject(r.Context(), projectName)
+	if err != nil {
+		return nil, err
+	}
+	return &project.NumericID, nil
+}
+
+func buildWebhookMetadata(in webhookThoughtRequest, headerKey string, now time.Time) thoughttypes.WebhookMetadata {
+	sourceMetadata := in.SourceMetadata
+	if len(sourceMetadata) == 0 {
+		sourceMetadata = in.Metadata
+	}
+	return thoughttypes.WebhookMetadata{
+		ReceivedAt:     now.Format(time.RFC3339),
+		IDempotencyKey: firstNonEmpty(in.IDempotencyKey, headerKey),
+		ExternalID:     strings.TrimSpace(in.ExternalID),
+		SourceMetadata: sanitizeWebhookMetadata(sourceMetadata),
+	}
+}
+
+func normalizeWebhookThoughtMetadata(in webhookThoughtRequest, webhookMeta thoughttypes.WebhookMetadata, capture config.CaptureConfig) thoughttypes.ThoughtMetadata {
+	return metadata.Normalize(thoughttypes.ThoughtMetadata{
+		People:         in.People,
+		ActionItems:    in.ActionItems,
+		DatesMentioned: in.DatesMentioned,
+		Topics:         in.Topics,
+		Type:           in.Type,
+		Source:         firstNonEmpty(in.Source, "webhook"),
+		Webhook:        &webhookMeta,
+	}, capture)
+}
+
+func sanitizeWebhookMetadata(in map[string]any) map[string]any {
+	if len(in) == 0 {
+		return nil
+	}
+	out := make(map[string]any, len(in))
+	for key, value := range in {
+		key = strings.TrimSpace(key)
+		if key == "" {
+			continue
+		}
+		if sanitized, ok := sanitizeWebhookMetadataValue(value, 0); ok {
+			out[key] = sanitized
+		}
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
+func sanitizeWebhookMetadataValue(value any, depth int) (any, bool) {
+	if depth > 3 {
+		return nil, false
+	}
+	switch v := value.(type) {
+	case nil, bool, float64, string:
+		return v, true
+	case []any:
+		if len(v) > 50 {
+			v = v[:50]
+		}
+		out := make([]any, 0, len(v))
+		for _, item := range v {
+			if sanitized, ok := sanitizeWebhookMetadataValue(item, depth+1); ok {
+				out = append(out, sanitized)
+			}
+		}
+		return out, true
+	case map[string]any:
+		if len(v) > 50 {
+			return nil, false
+		}
+		return sanitizeWebhookMetadata(v), true
+	default:
+		return nil, false
+	}
+}
+
+func writeWebhookThoughtResponse(w http.ResponseWriter, status int, out webhookThoughtResponse) {
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(status)
+	_ = json.NewEncoder(w).Encode(out)
+}
@@ -0,0 +1,86 @@
+package app
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"git.warky.dev/wdevs/amcs/internal/config"
+)
+
+func TestParseWebhookThoughtRequestRequiresJSON(t *testing.T) {
+	req := httptestRequest("text/plain", `{"content":"hello"}`)
+
+	_, err := parseWebhookThoughtRequest(req)
+	if err == nil {
+		t.Fatal("expected error for non-json content type")
+	}
+}
+
+func TestParseWebhookThoughtRequestRequiresContent(t *testing.T) {
+	req := httptestRequest("application/json", `{"source":"n8n"}`)
+
+	_, err := parseWebhookThoughtRequest(req)
+	if err == nil || !strings.Contains(err.Error(), "content is required") {
+		t.Fatalf("error = %v, want content required", err)
+	}
+}
+
+func TestBuildWebhookMetadataUsesHeaderIdempotencyAndSanitizesMetadata(t *testing.T) {
+	now := time.Date(2026, 7, 15, 4, 0, 0, 0, time.UTC)
+	got := buildWebhookMetadata(webhookThoughtRequest{
+		ExternalID: " ext-1 ",
+		Metadata: map[string]any{
+			"service": "n8n",
+			"unsafe":  struct{}{},
+			"nested":  map[string]any{"ok": true},
+		},
+	}, " key-1 ", now)
+
+	if got.IDempotencyKey != "key-1" {
+		t.Fatalf("IDempotencyKey = %q, want key-1", got.IDempotencyKey)
+	}
+	if got.ExternalID != "ext-1" {
+		t.Fatalf("ExternalID = %q, want ext-1", got.ExternalID)
+	}
+	if got.ReceivedAt != "2026-07-15T04:00:00Z" {
+		t.Fatalf("ReceivedAt = %q", got.ReceivedAt)
+	}
+	if _, ok := got.SourceMetadata["unsafe"]; ok {
+		t.Fatal("unsafe metadata value was not removed")
+	}
+	if got.SourceMetadata["service"] != "n8n" {
+		t.Fatalf("service metadata = %#v", got.SourceMetadata["service"])
+	}
+}
+
+func TestNormalizeWebhookThoughtMetadata(t *testing.T) {
+	webhookMeta := buildWebhookMetadata(webhookThoughtRequest{IDempotencyKey: "abc"}, "", time.Date(2026, 7, 15, 4, 0, 0, 0, time.UTC))
+	got := normalizeWebhookThoughtMetadata(webhookThoughtRequest{
+		Source: "github",
+		Type:   "task",
+		Topics: []string{"ci", "ci", ""},
+		People: []string{" Sam "},
+	}, webhookMeta, config.CaptureConfig{})
+
+	if got.Source != "github" {
+		t.Fatalf("Source = %q, want github", got.Source)
+	}
+	if got.Type != "task" {
+		t.Fatalf("Type = %q, want task", got.Type)
+	}
+	if len(got.Topics) != 1 || got.Topics[0] != "ci" {
+		t.Fatalf("Topics = %#v, want [ci]", got.Topics)
+	}
+	if got.Webhook == nil || got.Webhook.IDempotencyKey != "abc" {
+		t.Fatalf("Webhook = %#v, want idempotency key abc", got.Webhook)
+	}
+}
+
+func httptestRequest(contentType, body string) *http.Request {
+	req := httptest.NewRequest(http.MethodPost, "/webhooks/thoughts", strings.NewReader(body))
+	req.Header.Set("Content-Type", contentType)
+	return req
+}
@@ -53,6 +53,7 @@ func Normalize(in thoughttypes.ThoughtMetadata, capture config.CaptureConfig) th
 		Type:                    normalizeType(in.Type),
 		Source:                  normalizeSource(in.Source),
 		Attachments:             normalizeAttachments(in.Attachments),
+		Webhook:                 normalizeWebhook(in.Webhook),
 		MetadataStatus:          normalizeMetadataStatus(in.MetadataStatus),
 		MetadataUpdatedAt:       strings.TrimSpace(in.MetadataUpdatedAt),
 		MetadataLastAttemptedAt: strings.TrimSpace(in.MetadataLastAttemptedAt),
@@ -201,10 +202,31 @@ func Merge(base, patch thoughttypes.ThoughtMetadata, capture config.CaptureConfi
 	if len(patch.Attachments) > 0 {
 		merged.Attachments = append(append([]thoughttypes.ThoughtAttachment{}, merged.Attachments...), patch.Attachments...)
 	}
+	if patch.Webhook != nil {
+		merged.Webhook = patch.Webhook
+	}

 	return Normalize(merged, capture)
 }

+func normalizeWebhook(value *thoughttypes.WebhookMetadata) *thoughttypes.WebhookMetadata {
+	if value == nil {
+		return nil
+	}
+	out := &thoughttypes.WebhookMetadata{
+		ReceivedAt:     strings.TrimSpace(value.ReceivedAt),
+		IDempotencyKey: strings.TrimSpace(value.IDempotencyKey),
+		ExternalID:     strings.TrimSpace(value.ExternalID),
+	}
+	if len(value.SourceMetadata) > 0 {
+		out.SourceMetadata = value.SourceMetadata
+	}
+	if out.ReceivedAt == "" && out.IDempotencyKey == "" && out.ExternalID == "" && len(out.SourceMetadata) == 0 {
+		return nil
+	}
+	return out
+}
+
 func normalizeAttachments(values []thoughttypes.ThoughtAttachment) []thoughttypes.ThoughtAttachment {
 	seen := make(map[string]struct{}, len(values))
 	result := make([]thoughttypes.ThoughtAttachment, 0, len(values))
@@ -68,6 +68,22 @@ func (db *DB) InsertThought(ctx context.Context, thought thoughttypes.Thought, e
 	return created, nil
 }

+func (db *DB) GetThoughtByWebhookIDempotencyKey(ctx context.Context, key string) (thoughttypes.Thought, error) {
+	row := db.pool.QueryRow(ctx, `
+		select id, guid, content, metadata, project_id, archived_at, created_at, updated_at
+		from thoughts
+		where metadata->'webhook'->>'idempotency_key' = $1
+		order by created_at desc
+		limit 1
+	`, strings.TrimSpace(key))
+
+	var model generatedmodels.ModelPublicThoughts
+	if err := row.Scan(&model.ID, &model.GUID, &model.Content, &model.Metadata, &model.ProjectID, &model.ArchivedAt, &model.CreatedAt, &model.UpdatedAt); err != nil {
+		return thoughttypes.Thought{}, err
+	}
+	return thoughtFromModel(model)
+}
+
 func (db *DB) SearchThoughts(ctx context.Context, embedding []float32, embeddingModel string, threshold float64, limit int, filter map[string]any) ([]thoughttypes.SearchResult, error) {
 	filterJSON, err := json.Marshal(filter)
 	if err != nil {
@@ -39,6 +39,7 @@ type ProjectContextOutput struct {
 	Project       thoughttypes.Project `json:"project"`
 	Context       string               `json:"context"`
 	Items         []ContextItem        `json:"items"`
+	RetrievalMode string               `json:"retrieval_mode,omitempty"`
 }

 func NewContextTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *ContextTool {
@@ -70,12 +71,14 @@ func (t *ContextTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in P
 		})
 	}

+	var retrievalMode string
 	query := strings.TrimSpace(in.Query)
 	if query != "" {
-		semantic, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, &project.NumericID, nil)
+		semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, &project.NumericID, nil)
 		if err != nil {
 			return nil, ProjectContextOutput{}, err
 		}
+		retrievalMode = mode
 		for _, result := range semantic {
 			key := fmt.Sprint(result.ID)
 			if _, ok := seen[key]; ok {
@@ -103,5 +106,6 @@ func (t *ContextTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in P
 		Project:       *project,
 		Context:       contextBlock,
 		Items:         items,
+		RetrievalMode: retrievalMode,
 	}, nil
 }
@@ -46,6 +46,7 @@ type RelatedThought struct {

 type RelatedOutput struct {
 	Related       []RelatedThought `json:"related"`
+	RetrievalMode string           `json:"retrieval_mode,omitempty"`
 }

 func NewLinksTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig) *LinksTool {
@@ -117,11 +118,13 @@ func (t *LinksTool) Related(ctx context.Context, _ *mcp.CallToolRequest, in Rela
 		includeSemantic = *in.IncludeSemantic
 	}

+	var retrievalMode string
 	if includeSemantic {
-		semantic, err := semanticSearch(ctx, t.store, t.embeddings, t.search, thought.Content, t.search.DefaultLimit, t.search.DefaultThreshold, thought.ProjectID, &thought.GUID)
+		semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, thought.Content, t.search.DefaultLimit, t.search.DefaultThreshold, thought.ProjectID, &thought.GUID)
 		if err != nil {
 			return nil, RelatedOutput{}, err
 		}
+		retrievalMode = mode
 		for _, item := range semantic {
 			key := fmt.Sprint(item.ID)
 			if _, ok := seen[key]; ok {
@@ -138,5 +141,5 @@ func (t *LinksTool) Related(ctx context.Context, _ *mcp.CallToolRequest, in Rela
 		}
 	}

-	return nil, RelatedOutput{Related: related}, nil
+	return nil, RelatedOutput{Related: related, RetrievalMode: retrievalMode}, nil
 }
@@ -29,6 +29,7 @@ type RecallInput struct {
 type RecallOutput struct {
 	Context       string        `json:"context"`
 	Items         []ContextItem `json:"items"`
+	RetrievalMode string        `json:"retrieval_mode,omitempty"`
 }

 func NewRecallTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *RecallTool {
@@ -53,7 +54,7 @@ func (t *RecallTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Re
 		projectID = &project.NumericID
 	}

-	semantic, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
+	semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
 	if err != nil {
 		return nil, RecallOutput{}, err
 	}
@@ -102,5 +103,6 @@ func (t *RecallTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Re
 	return nil, RecallOutput{
 		Context:       formatContextBlock(header, lines),
 		Items:         items,
+		RetrievalMode: mode,
 	}, nil
 }
@@ -11,10 +11,16 @@ import (
 	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
 )

+const (
+	RetrievalModeSemantic = "semantic"
+	RetrievalModeText     = "text"
+)
+
 // semanticSearch runs vector similarity search if embeddings exist for the
 // primary embedding model in the given scope, otherwise falls back to Postgres
 // full-text search. Search always uses the primary model so query vectors
 // match rows stored under the primary model name.
+// It returns the results and the retrieval mode used ("semantic" or "text").
 func semanticSearch(
 	ctx context.Context,
 	db *store.DB,
@@ -25,20 +31,22 @@ func semanticSearch(
 	threshold float64,
 	projectID *int64,
 	excludeID *uuid.UUID,
-) ([]thoughttypes.SearchResult, error) {
+) ([]thoughttypes.SearchResult, string, error) {
 	model := embeddings.PrimaryModel()
 	hasEmbeddings, err := db.HasEmbeddingsForModel(ctx, model, projectID)
 	if err != nil {
-		return nil, err
+		return nil, "", err
 	}

 	if hasEmbeddings {
 		embedding, err := embeddings.EmbedPrimary(ctx, query)
 		if err != nil {
-			return nil, err
+			return nil, "", err
 		}
-		return db.SearchSimilarThoughts(ctx, embedding, model, threshold, limit, projectID, excludeID)
+		results, err := db.SearchSimilarThoughts(ctx, embedding, model, threshold, limit, projectID, excludeID)
+		return results, RetrievalModeSemantic, err
 	}

-	return db.SearchThoughtsText(ctx, query, limit, projectID, excludeID)
+	results, err := db.SearchThoughtsText(ctx, query, limit, projectID, excludeID)
+	return results, RetrievalModeText, err
 }
@@ -0,0 +1,60 @@
+package tools
+
+import "testing"
+
+func TestRetrievalModeConstants(t *testing.T) {
+	if RetrievalModeSemantic != "semantic" {
+		t.Fatalf("RetrievalModeSemantic = %q, want %q", RetrievalModeSemantic, "semantic")
+	}
+	if RetrievalModeText != "text" {
+		t.Fatalf("RetrievalModeText = %q, want %q", RetrievalModeText, "text")
+	}
+}
+
+func TestSearchOutputIncludesRetrievalMode(t *testing.T) {
+	out := SearchOutput{RetrievalMode: RetrievalModeSemantic}
+	if out.RetrievalMode != RetrievalModeSemantic {
+		t.Fatalf("SearchOutput.RetrievalMode = %q, want %q", out.RetrievalMode, RetrievalModeSemantic)
+	}
+}
+
+func TestRelatedOutputIncludesRetrievalMode(t *testing.T) {
+	out := RelatedOutput{RetrievalMode: RetrievalModeText}
+	if out.RetrievalMode != RetrievalModeText {
+		t.Fatalf("RelatedOutput.RetrievalMode = %q, want %q", out.RetrievalMode, RetrievalModeText)
+	}
+	// retrieval_mode is omitempty — empty string means semantic search was skipped
+	out2 := RelatedOutput{}
+	if out2.RetrievalMode != "" {
+		t.Fatalf("RelatedOutput.RetrievalMode = %q, want empty when include_semantic is false", out2.RetrievalMode)
+	}
+}
+
+func TestSummarizeOutputIncludesRetrievalModeOnlyWhenQueryUsed(t *testing.T) {
+	withQuery := SummarizeOutput{Summary: "s", Count: 1, RetrievalMode: RetrievalModeSemantic}
+	if withQuery.RetrievalMode != RetrievalModeSemantic {
+		t.Fatalf("SummarizeOutput.RetrievalMode = %q, want %q", withQuery.RetrievalMode, RetrievalModeSemantic)
+	}
+	withoutQuery := SummarizeOutput{Summary: "s", Count: 1}
+	if withoutQuery.RetrievalMode != "" {
+		t.Fatalf("SummarizeOutput.RetrievalMode = %q, want empty when no query", withoutQuery.RetrievalMode)
+	}
+}
+
+func TestProjectContextOutputIncludesRetrievalModeOnlyWhenQueryUsed(t *testing.T) {
+	withQuery := ProjectContextOutput{RetrievalMode: RetrievalModeText}
+	if withQuery.RetrievalMode != RetrievalModeText {
+		t.Fatalf("ProjectContextOutput.RetrievalMode = %q, want %q", withQuery.RetrievalMode, RetrievalModeText)
+	}
+	withoutQuery := ProjectContextOutput{}
+	if withoutQuery.RetrievalMode != "" {
+		t.Fatalf("ProjectContextOutput.RetrievalMode = %q, want empty when no query", withoutQuery.RetrievalMode)
+	}
+}
+
+func TestRecallOutputIncludesRetrievalMode(t *testing.T) {
+	out := RecallOutput{RetrievalMode: RetrievalModeSemantic}
+	if out.RetrievalMode != RetrievalModeSemantic {
+		t.Fatalf("RecallOutput.RetrievalMode = %q, want %q", out.RetrievalMode, RetrievalModeSemantic)
+	}
+}
@@ -29,6 +29,7 @@ type SearchInput struct {

 type SearchOutput struct {
 	Results       []thoughttypes.SearchResult `json:"results"`
+	RetrievalMode string                      `json:"retrieval_mode,omitempty"`
 }

 func NewSearchTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *SearchTool {
@@ -55,10 +56,10 @@ func (t *SearchTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Se
 		_ = t.store.TouchProject(ctx, project.NumericID)
 	}

-	results, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, threshold, projectID, nil)
+	results, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, threshold, projectID, nil)
 	if err != nil {
 		return nil, SearchOutput{}, err
 	}

-	return nil, SearchOutput{Results: results}, nil
+	return nil, SearchOutput{Results: results, RetrievalMode: mode}, nil
 }
@@ -30,6 +30,7 @@ type SummarizeInput struct {
 type SummarizeOutput struct {
 	Summary       string `json:"summary"`
 	Count         int    `json:"count"`
+	RetrievalMode string `json:"retrieval_mode,omitempty"`
 }

 func NewSummarizeTool(db *store.DB, embeddings *ai.EmbeddingRunner, metadata *ai.MetadataRunner, search config.SearchConfig, sessions *session.ActiveProjects) *SummarizeTool {
@@ -47,15 +48,17 @@ func (t *SummarizeTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in
 	lines := make([]string, 0, limit)
 	count := 0

+	var retrievalMode string
 	if query != "" {
 		var projectID *int64
 		if project != nil {
 			projectID = &project.NumericID
 		}
-		results, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
+		results, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
 		if err != nil {
 			return nil, SummarizeOutput{}, err
 		}
+		retrievalMode = mode
 		for i, result := range results {
 			lines = append(lines, thoughtContextLine(i, result.Content, result.Metadata, result.Similarity))
 		}
@@ -85,5 +88,5 @@ func (t *SummarizeTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in
 		_ = t.store.TouchProject(ctx, project.NumericID)
 	}

-	return nil, SummarizeOutput{Summary: summary, Count: count}, nil
+	return nil, SummarizeOutput{Summary: summary, Count: count, RetrievalMode: retrievalMode}, nil
 }
@@ -14,12 +14,20 @@ type ThoughtMetadata struct {
 	Type                    string              `json:"type"`
 	Source                  string              `json:"source"`
 	Attachments             []ThoughtAttachment `json:"attachments,omitempty"`
+	Webhook                 *WebhookMetadata    `json:"webhook,omitempty"`
 	MetadataStatus          string              `json:"metadata_status,omitempty"`
 	MetadataUpdatedAt       string              `json:"metadata_updated_at,omitempty"`
 	MetadataLastAttemptedAt string              `json:"metadata_last_attempted_at,omitempty"`
 	MetadataError           string              `json:"metadata_error,omitempty"`
 }

+type WebhookMetadata struct {
+	ReceivedAt     string         `json:"received_at"`
+	IDempotencyKey string         `json:"idempotency_key,omitempty"`
+	ExternalID     string         `json:"external_id,omitempty"`
+	SourceMetadata map[string]any `json:"source_metadata,omitempty"`
+}
+
 type ThoughtAttachment struct {
 	FileID    uuid.UUID `json:"file_id"`
 	Name      string    `json:"name"`