Author SHA1 Message Date
warkanum 8db2141d45 ci: build ui before go tests
CI / build-and-test (push) Successful in 7m10s
CI / build-and-test (pull_request) Successful in 7m13s
2026-07-14 16:32:00 +02:00
warkanum 3198600031 feat(tools): add duplicate audit report
CI / build-and-test (push) Failing after 1m24s
CI / build-and-test (pull_request) Failing after 59s
2026-07-14 16:28:06 +02:00
13 changed files with 450 additions and 127 deletions
+14
View File
@@ -31,6 +31,20 @@ jobs:
- name: Download dependencies
run: go mod download
- name: Set up Node
uses: actions/setup-node@v4
with:
node-version: 'lts/*'
- name: Install pnpm
run: npm install -g pnpm
- name: Build UI
run: |
cd ui
pnpm install --frozen-lockfile
pnpm run build
- name: Tidy modules
run: go mod tidy
+1 -1
View File
@@ -478,7 +478,7 @@ 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. All five tools include a `retrieval_mode` field in their response (`"semantic"` or `"text"`) so callers can see which path was taken.
**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.
## Client Setup
+1
View File
@@ -202,6 +202,7 @@ func routes(logger *slog.Logger, cfg *config.Config, info buildinfo.Info, db *st
Update: tools.NewUpdateTool(db, embeddings, metadata, cfg.Capture, logger),
Delete: tools.NewDeleteTool(db),
Archive: tools.NewArchiveTool(db),
DuplicateAudit: tools.NewDuplicateAuditTool(db, cfg.Search, activeProjects),
Projects: tools.NewProjectsTool(db, activeProjects),
Version: tools.NewVersionTool(cfg.MCP.ServerName, info),
Learnings: tools.NewLearningsTool(db, activeProjects, cfg.Search),
+26 -18
View File
@@ -19,24 +19,25 @@ const (
)
type ToolSet struct {
Version *tools.VersionTool
Capture *tools.CaptureTool
Search *tools.SearchTool
List *tools.ListTool
Stats *tools.StatsTool
Get *tools.GetTool
Update *tools.UpdateTool
Delete *tools.DeleteTool
Archive *tools.ArchiveTool
Projects *tools.ProjectsTool
Context *tools.ContextTool
Recall *tools.RecallTool
Summarize *tools.SummarizeTool
Links *tools.LinksTool
Files *tools.FilesTool
Backfill *tools.BackfillTool
Reparse *tools.ReparseMetadataTool
RetryMetadata *tools.RetryEnrichmentTool
Version *tools.VersionTool
Capture *tools.CaptureTool
Search *tools.SearchTool
List *tools.ListTool
Stats *tools.StatsTool
Get *tools.GetTool
Update *tools.UpdateTool
Delete *tools.DeleteTool
Archive *tools.ArchiveTool
DuplicateAudit *tools.DuplicateAuditTool
Projects *tools.ProjectsTool
Context *tools.ContextTool
Recall *tools.RecallTool
Summarize *tools.SummarizeTool
Links *tools.LinksTool
Files *tools.FilesTool
Backfill *tools.BackfillTool
Reparse *tools.ReparseMetadataTool
RetryMetadata *tools.RetryEnrichmentTool
//Maintenance *tools.MaintenanceTool
Skills *tools.SkillsTool
Personas *tools.AgentPersonasTool
@@ -227,6 +228,12 @@ func registerThoughtTools(server *mcp.Server, logger *slog.Logger, toolSet ToolS
}, toolSet.Archive.Handle); err != nil {
return err
}
if err := addTool(server, logger, &mcp.Tool{
Name: "audit_duplicates",
Description: "Dry-run duplicate audit for projects, thoughts, and metadata normalization candidates; performs no writes.",
}, toolSet.DuplicateAudit.Handle); err != nil {
return err
}
if err := addTool(server, logger, &mcp.Tool{
Name: "summarize_thoughts",
Description: "LLM summary of a filtered set of thoughts.",
@@ -764,6 +771,7 @@ func BuildToolCatalog() []tools.ToolEntry {
{Name: "update_thought", Description: "Update thought content or merge metadata.", Category: "thoughts"},
{Name: "delete_thought", Description: "Hard-delete a thought by id.", Category: "thoughts"},
{Name: "archive_thought", Description: "Archive a thought so it is hidden from default search and listing.", Category: "thoughts"},
{Name: "audit_duplicates", Description: "Dry-run duplicate audit for exact/normalized project names, thought content, and metadata value variants. Reports candidates and recommendations; performs no writes.", Category: "admin"},
{Name: "summarize_thoughts", Description: "Produce an LLM prose summary of a filtered or searched set of thoughts.", Category: "thoughts"},
{Name: "recall_context", Description: "Recall semantically relevant and recent context for prompt injection. Combines vector similarity with recency. Falls back to full-text search when no embeddings exist.", Category: "thoughts"},
{Name: "link_thoughts", Description: "Create a typed relationship between two thoughts.", Category: "thoughts"},
+7 -11
View File
@@ -36,10 +36,9 @@ type ContextItem struct {
}
type ProjectContextOutput struct {
Project thoughttypes.Project `json:"project"`
Context string `json:"context"`
Items []ContextItem `json:"items"`
RetrievalMode string `json:"retrieval_mode,omitempty"`
Project thoughttypes.Project `json:"project"`
Context string `json:"context"`
Items []ContextItem `json:"items"`
}
func NewContextTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *ContextTool {
@@ -71,14 +70,12 @@ func (t *ContextTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in P
})
}
var retrievalMode string
query := strings.TrimSpace(in.Query)
if query != "" {
semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, &project.NumericID, nil)
semantic, 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 {
@@ -103,9 +100,8 @@ func (t *ContextTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in P
_ = t.store.TouchProject(ctx, project.NumericID)
return nil, ProjectContextOutput{
Project: *project,
Context: contextBlock,
Items: items,
RetrievalMode: retrievalMode,
Project: *project,
Context: contextBlock,
Items: items,
}, nil
}
+305
View File
@@ -0,0 +1,305 @@
package tools
import (
"context"
"regexp"
"sort"
"strings"
"github.com/google/uuid"
"github.com/modelcontextprotocol/go-sdk/mcp"
"git.warky.dev/wdevs/amcs/internal/config"
"git.warky.dev/wdevs/amcs/internal/session"
"git.warky.dev/wdevs/amcs/internal/store"
thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
)
const defaultDuplicateAuditLimit = 50
var duplicateWhitespace = regexp.MustCompile(`\s+`)
type DuplicateAuditTool struct {
store *store.DB
search config.SearchConfig
sessions *session.ActiveProjects
}
type DuplicateAuditInput struct {
Project string `json:"project,omitempty" jsonschema:"optional project name or id to scope thought duplicate scanning"`
Limit int `json:"limit,omitempty" jsonschema:"maximum candidate groups to return"`
ThoughtLimit int `json:"thought_limit,omitempty" jsonschema:"maximum thoughts to scan for duplicate content; defaults to search limit"`
IncludeArchived bool `json:"include_archived,omitempty" jsonschema:"include archived thoughts in the scan"`
}
type DuplicateAuditOutput struct {
Report DuplicateAuditReport `json:"report"`
}
type DuplicateAuditReport struct {
Summary DuplicateAuditSummary `json:"summary"`
ProjectCandidates []DuplicateProjectCandidate `json:"project_candidates"`
ThoughtCandidates []DuplicateThoughtCandidate `json:"thought_candidates"`
MetadataCandidates []MetadataCleanupCandidate `json:"metadata_candidates"`
RecommendedNextStep string `json:"recommended_next_step"`
}
type DuplicateAuditSummary struct {
DryRun bool `json:"dry_run"`
ScannedProjects int `json:"scanned_projects"`
ScannedThoughts int `json:"scanned_thoughts"`
ProjectCandidateGroups int `json:"project_candidate_groups"`
ThoughtCandidateGroups int `json:"thought_candidate_groups"`
MetadataCandidateGroups int `json:"metadata_candidate_groups"`
Truncated bool `json:"truncated"`
}
type DuplicateProjectCandidate struct {
MatchType string `json:"match_type"`
NormalizedValue string `json:"normalized_value"`
Confidence float64 `json:"confidence"`
RecommendedCanonicalID uuid.UUID `json:"recommended_canonical_id"`
Projects []thoughttypes.ProjectSummary `json:"projects"`
}
type DuplicateThoughtCandidate struct {
MatchType string `json:"match_type"`
NormalizedValue string `json:"normalized_value"`
Confidence float64 `json:"confidence"`
RecommendedCanonicalID uuid.UUID `json:"recommended_canonical_id"`
Thoughts []thoughttypes.Thought `json:"thoughts"`
}
type MetadataCleanupCandidate struct {
Field string `json:"field"`
NormalizedValue string `json:"normalized_value"`
Values []string `json:"values"`
Occurrences int `json:"occurrences"`
RecommendedValue string `json:"recommended_value"`
RecommendedAction string `json:"recommended_action"`
}
func NewDuplicateAuditTool(db *store.DB, search config.SearchConfig, sessions *session.ActiveProjects) *DuplicateAuditTool {
return &DuplicateAuditTool{store: db, search: search, sessions: sessions}
}
func (t *DuplicateAuditTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in DuplicateAuditInput) (*mcp.CallToolResult, DuplicateAuditOutput, error) {
project, err := resolveProject(ctx, t.store, t.sessions, req, in.Project, false)
if err != nil {
return nil, DuplicateAuditOutput{}, err
}
var projectID *int64
if project != nil {
projectID = &project.NumericID
}
projects, err := t.store.ListProjects(ctx)
if err != nil {
return nil, DuplicateAuditOutput{}, err
}
thoughtLimit := normalizeLimit(in.ThoughtLimit, t.search)
if in.ThoughtLimit <= 0 {
thoughtLimit = normalizeLimit(0, t.search)
}
thoughts, err := t.store.ListThoughts(ctx, thoughttypes.ListFilter{Limit: thoughtLimit, ProjectID: projectID, IncludeArchived: in.IncludeArchived})
if err != nil {
return nil, DuplicateAuditOutput{}, err
}
if project != nil {
_ = t.store.TouchProject(ctx, project.NumericID)
}
return nil, DuplicateAuditOutput{Report: buildDuplicateAuditReport(projects, thoughts, in)}, nil
}
func buildDuplicateAuditReport(projects []thoughttypes.ProjectSummary, thoughts []thoughttypes.Thought, in DuplicateAuditInput) DuplicateAuditReport {
limit := in.Limit
if limit <= 0 {
limit = defaultDuplicateAuditLimit
}
projectCandidates, projectTruncated := duplicateProjectCandidates(projects, limit)
thoughtCandidates, thoughtTruncated := duplicateThoughtCandidates(thoughts, limit)
metadataCandidates, metadataTruncated := metadataCleanupCandidates(thoughts, limit)
return DuplicateAuditReport{
Summary: DuplicateAuditSummary{
DryRun: true,
ScannedProjects: len(projects),
ScannedThoughts: len(thoughts),
ProjectCandidateGroups: len(projectCandidates),
ThoughtCandidateGroups: len(thoughtCandidates),
MetadataCandidateGroups: len(metadataCandidates),
Truncated: projectTruncated || thoughtTruncated || metadataTruncated,
},
ProjectCandidates: projectCandidates,
ThoughtCandidates: thoughtCandidates,
MetadataCandidates: metadataCandidates,
RecommendedNextStep: "Review candidates manually, then use explicit archive/update/merge tools; this audit performs no writes.",
}
}
func duplicateProjectCandidates(projects []thoughttypes.ProjectSummary, limit int) ([]DuplicateProjectCandidate, bool) {
groups := map[string][]thoughttypes.ProjectSummary{}
for _, project := range projects {
key := normalizeDuplicateText(project.Name)
if key != "" {
groups[key] = append(groups[key], project)
}
}
keys := sortedDuplicateKeys(groups)
out := make([]DuplicateProjectCandidate, 0, min(len(keys), limit))
for _, key := range keys {
items := groups[key]
sort.SliceStable(items, func(i, j int) bool {
if items[i].ThoughtCount != items[j].ThoughtCount {
return items[i].ThoughtCount > items[j].ThoughtCount
}
return items[i].CreatedAt.Before(items[j].CreatedAt)
})
matchType := "normalized_name"
if allProjectNamesExact(items) {
matchType = "exact_name"
}
out = append(out, DuplicateProjectCandidate{MatchType: matchType, NormalizedValue: key, Confidence: 1.0, RecommendedCanonicalID: items[0].ID, Projects: items})
if len(out) == limit {
return out, len(keys) > limit
}
}
return out, false
}
func duplicateThoughtCandidates(thoughts []thoughttypes.Thought, limit int) ([]DuplicateThoughtCandidate, bool) {
groups := map[string][]thoughttypes.Thought{}
for _, thought := range thoughts {
key := normalizeDuplicateText(thought.Content)
if key != "" {
groups[key] = append(groups[key], thought)
}
}
keys := sortedDuplicateKeys(groups)
out := make([]DuplicateThoughtCandidate, 0, min(len(keys), limit))
for _, key := range keys {
items := groups[key]
sort.SliceStable(items, func(i, j int) bool { return items[i].CreatedAt.Before(items[j].CreatedAt) })
matchType := "normalized_content"
if allThoughtContentExact(items) {
matchType = "exact_content"
}
out = append(out, DuplicateThoughtCandidate{MatchType: matchType, NormalizedValue: key, Confidence: 1.0, RecommendedCanonicalID: items[0].GUID, Thoughts: items})
if len(out) == limit {
return out, len(keys) > limit
}
}
return out, false
}
func metadataCleanupCandidates(thoughts []thoughttypes.Thought, limit int) ([]MetadataCleanupCandidate, bool) {
groups := map[string]map[string]int{}
add := func(field, value string) {
trimmed := strings.TrimSpace(value)
key := normalizeDuplicateText(trimmed)
if key == "" || trimmed == "" {
return
}
bucketKey := field + "\x00" + key
if groups[bucketKey] == nil {
groups[bucketKey] = map[string]int{}
}
groups[bucketKey][trimmed]++
}
for _, thought := range thoughts {
add("type", thought.Metadata.Type)
add("source", thought.Metadata.Source)
for _, topic := range thought.Metadata.Topics {
add("topics", topic)
}
for _, person := range thought.Metadata.People {
add("people", person)
}
}
keys := make([]string, 0, len(groups))
for key, values := range groups {
if len(values) > 1 {
keys = append(keys, key)
}
}
sort.Strings(keys)
out := make([]MetadataCleanupCandidate, 0, min(len(keys), limit))
for _, key := range keys {
parts := strings.SplitN(key, "\x00", 2)
values, occurrences, recommended := valuesByCount(groups[key])
out = append(out, MetadataCleanupCandidate{
Field: parts[0],
NormalizedValue: parts[1],
Values: values,
Occurrences: occurrences,
RecommendedValue: recommended,
RecommendedAction: "preview bulk metadata remap before applying; no changes made by audit",
})
if len(out) == limit {
return out, len(keys) > limit
}
}
return out, false
}
func sortedDuplicateKeys[T any](groups map[string][]T) []string {
keys := make([]string, 0, len(groups))
for key, items := range groups {
if len(items) > 1 {
keys = append(keys, key)
}
}
sort.Strings(keys)
return keys
}
func valuesByCount(counts map[string]int) ([]string, int, string) {
values := make([]string, 0, len(counts))
occurrences := 0
for value, count := range counts {
values = append(values, value)
occurrences += count
}
sort.Slice(values, func(i, j int) bool {
if counts[values[i]] != counts[values[j]] {
return counts[values[i]] > counts[values[j]]
}
return values[i] < values[j]
})
return values, occurrences, values[0]
}
func normalizeDuplicateText(value string) string {
return duplicateWhitespace.ReplaceAllString(strings.ToLower(strings.TrimSpace(value)), " ")
}
func allProjectNamesExact(projects []thoughttypes.ProjectSummary) bool {
if len(projects) == 0 {
return false
}
first := strings.TrimSpace(projects[0].Name)
for _, project := range projects[1:] {
if strings.TrimSpace(project.Name) != first {
return false
}
}
return true
}
func allThoughtContentExact(thoughts []thoughttypes.Thought) bool {
if len(thoughts) == 0 {
return false
}
first := strings.TrimSpace(thoughts[0].Content)
for _, thought := range thoughts[1:] {
if strings.TrimSpace(thought.Content) != first {
return false
}
}
return true
}
+76
View File
@@ -0,0 +1,76 @@
package tools
import (
"testing"
"time"
"github.com/google/uuid"
thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
)
func TestBuildDuplicateAuditGroupsProjectsAndThoughtsSafely(t *testing.T) {
now := time.Date(2026, 4, 12, 20, 5, 0, 0, time.UTC)
projects := []thoughttypes.ProjectSummary{
{Project: thoughttypes.Project{ID: uuid.MustParse("11111111-1111-1111-1111-111111111111"), NumericID: 1, Name: "AMCS", CreatedAt: now}, ThoughtCount: 3},
{Project: thoughttypes.Project{ID: uuid.MustParse("22222222-2222-2222-2222-222222222222"), NumericID: 2, Name: "amcs ", CreatedAt: now.Add(time.Hour)}, ThoughtCount: 1},
{Project: thoughttypes.Project{ID: uuid.MustParse("33333333-3333-3333-3333-333333333333"), NumericID: 3, Name: "other", CreatedAt: now}, ThoughtCount: 1},
}
thoughts := []thoughttypes.Thought{
{ID: 10, GUID: uuid.MustParse("aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa"), Content: "Same text", Metadata: thoughttypes.ThoughtMetadata{Type: "Note", Topics: []string{"Go", "go"}, People: []string{"Alice"}}, CreatedAt: now},
{ID: 11, GUID: uuid.MustParse("bbbbbbbb-bbbb-bbbb-bbbb-bbbbbbbbbbbb"), Content: " same text ", Metadata: thoughttypes.ThoughtMetadata{Type: "note", Topics: []string{"go"}, People: []string{"alice"}}, CreatedAt: now.Add(time.Hour)},
{ID: 12, GUID: uuid.MustParse("cccccccc-cccc-cccc-cccc-cccccccccccc"), Content: "unique", Metadata: thoughttypes.ThoughtMetadata{Type: "Task", Topics: []string{"Tasks"}}, CreatedAt: now},
}
report := buildDuplicateAuditReport(projects, thoughts, DuplicateAuditInput{})
if len(report.ProjectCandidates) != 1 {
t.Fatalf("project candidates = %d, want 1: %#v", len(report.ProjectCandidates), report.ProjectCandidates)
}
projectCandidate := report.ProjectCandidates[0]
if projectCandidate.MatchType != "normalized_name" || projectCandidate.NormalizedValue != "amcs" || len(projectCandidate.Projects) != 2 {
t.Fatalf("project candidate mismatch: %#v", projectCandidate)
}
if projectCandidate.RecommendedCanonicalID != projects[0].ID {
t.Fatalf("recommended project = %s, want %s", projectCandidate.RecommendedCanonicalID, projects[0].ID)
}
if len(report.ThoughtCandidates) != 1 {
t.Fatalf("thought candidates = %d, want 1: %#v", len(report.ThoughtCandidates), report.ThoughtCandidates)
}
thoughtCandidate := report.ThoughtCandidates[0]
if thoughtCandidate.MatchType != "normalized_content" || thoughtCandidate.NormalizedValue != "same text" || len(thoughtCandidate.Thoughts) != 2 {
t.Fatalf("thought candidate mismatch: %#v", thoughtCandidate)
}
if thoughtCandidate.RecommendedCanonicalID != thoughts[0].GUID {
t.Fatalf("recommended thought = %s, want %s", thoughtCandidate.RecommendedCanonicalID, thoughts[0].GUID)
}
if len(report.MetadataCandidates) != 3 {
t.Fatalf("metadata candidates = %d, want 3: %#v", len(report.MetadataCandidates), report.MetadataCandidates)
}
if report.Summary.ProjectCandidateGroups != 1 || report.Summary.ThoughtCandidateGroups != 1 || report.Summary.MetadataCandidateGroups != 3 {
t.Fatalf("summary mismatch: %#v", report.Summary)
}
if report.Summary.DryRun != true {
t.Fatalf("dry run = false, want true")
}
}
func TestBuildDuplicateAuditRespectsLimit(t *testing.T) {
projects := []thoughttypes.ProjectSummary{
{Project: thoughttypes.Project{ID: uuid.MustParse("11111111-1111-1111-1111-111111111111"), Name: "Alpha"}},
{Project: thoughttypes.Project{ID: uuid.MustParse("22222222-2222-2222-2222-222222222222"), Name: "alpha"}},
{Project: thoughttypes.Project{ID: uuid.MustParse("33333333-3333-3333-3333-333333333333"), Name: "Beta"}},
{Project: thoughttypes.Project{ID: uuid.MustParse("44444444-4444-4444-4444-444444444444"), Name: "beta"}},
}
report := buildDuplicateAuditReport(projects, nil, DuplicateAuditInput{Limit: 1})
if len(report.ProjectCandidates) != 1 {
t.Fatalf("project candidates = %d, want 1", len(report.ProjectCandidates))
}
if report.Summary.Truncated != true {
t.Fatalf("truncated = false, want true")
}
}
+3 -6
View File
@@ -45,8 +45,7 @@ type RelatedThought struct {
}
type RelatedOutput struct {
Related []RelatedThought `json:"related"`
RetrievalMode string `json:"retrieval_mode,omitempty"`
Related []RelatedThought `json:"related"`
}
func NewLinksTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig) *LinksTool {
@@ -118,13 +117,11 @@ func (t *LinksTool) Related(ctx context.Context, _ *mcp.CallToolRequest, in Rela
includeSemantic = *in.IncludeSemantic
}
var retrievalMode string
if includeSemantic {
semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, thought.Content, t.search.DefaultLimit, t.search.DefaultThreshold, thought.ProjectID, &thought.GUID)
semantic, 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 {
@@ -141,5 +138,5 @@ func (t *LinksTool) Related(ctx context.Context, _ *mcp.CallToolRequest, in Rela
}
}
return nil, RelatedOutput{Related: related, RetrievalMode: retrievalMode}, nil
return nil, RelatedOutput{Related: related}, nil
}
+5 -7
View File
@@ -27,9 +27,8 @@ type RecallInput struct {
}
type RecallOutput struct {
Context string `json:"context"`
Items []ContextItem `json:"items"`
RetrievalMode string `json:"retrieval_mode,omitempty"`
Context string `json:"context"`
Items []ContextItem `json:"items"`
}
func NewRecallTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *RecallTool {
@@ -54,7 +53,7 @@ func (t *RecallTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Re
projectID = &project.NumericID
}
semantic, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
semantic, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
if err != nil {
return nil, RecallOutput{}, err
}
@@ -101,8 +100,7 @@ func (t *RecallTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Re
}
return nil, RecallOutput{
Context: formatContextBlock(header, lines),
Items: items,
RetrievalMode: mode,
Context: formatContextBlock(header, lines),
Items: items,
}, nil
}
+5 -13
View File
@@ -11,16 +11,10 @@ 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,
@@ -31,22 +25,20 @@ func semanticSearch(
threshold float64,
projectID *int64,
excludeID *uuid.UUID,
) ([]thoughttypes.SearchResult, string, error) {
) ([]thoughttypes.SearchResult, 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
}
results, err := db.SearchSimilarThoughts(ctx, embedding, model, threshold, limit, projectID, excludeID)
return results, RetrievalModeSemantic, err
return db.SearchSimilarThoughts(ctx, embedding, model, threshold, limit, projectID, excludeID)
}
results, err := db.SearchThoughtsText(ctx, query, limit, projectID, excludeID)
return results, RetrievalModeText, err
return db.SearchThoughtsText(ctx, query, limit, projectID, excludeID)
}
-60
View File
@@ -1,60 +0,0 @@
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)
}
}
+3 -4
View File
@@ -28,8 +28,7 @@ type SearchInput struct {
}
type SearchOutput struct {
Results []thoughttypes.SearchResult `json:"results"`
RetrievalMode string `json:"retrieval_mode,omitempty"`
Results []thoughttypes.SearchResult `json:"results"`
}
func NewSearchTool(db *store.DB, embeddings *ai.EmbeddingRunner, search config.SearchConfig, sessions *session.ActiveProjects) *SearchTool {
@@ -56,10 +55,10 @@ func (t *SearchTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in Se
_ = t.store.TouchProject(ctx, project.NumericID)
}
results, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, threshold, projectID, nil)
results, 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, RetrievalMode: mode}, nil
return nil, SearchOutput{Results: results}, nil
}
+4 -7
View File
@@ -28,9 +28,8 @@ type SummarizeInput struct {
}
type SummarizeOutput struct {
Summary string `json:"summary"`
Count int `json:"count"`
RetrievalMode string `json:"retrieval_mode,omitempty"`
Summary string `json:"summary"`
Count int `json:"count"`
}
func NewSummarizeTool(db *store.DB, embeddings *ai.EmbeddingRunner, metadata *ai.MetadataRunner, search config.SearchConfig, sessions *session.ActiveProjects) *SummarizeTool {
@@ -48,17 +47,15 @@ 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, mode, err := semanticSearch(ctx, t.store, t.embeddings, t.search, query, limit, t.search.DefaultThreshold, projectID, nil)
results, 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))
}
@@ -88,5 +85,5 @@ func (t *SummarizeTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in
_ = t.store.TouchProject(ctx, project.NumericID)
}
return nil, SummarizeOutput{Summary: summary, Count: count, RetrievalMode: retrievalMode}, nil
return nil, SummarizeOutput{Summary: summary, Count: count}, nil
}