diff --git a/internal/runtime/memory/projectindex/chapter.go b/internal/runtime/memory/projectindex/chapter.go
new file mode 100644
index 0000000..f5c49d9
--- /dev/null
+++ b/internal/runtime/memory/projectindex/chapter.go
@@ -0,0 +1,110 @@
+package projectindex
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"time"
+)
+
+// Chapter is a project chapter (layer 3 of the hierarchy): the deep content a
+// project is split into. ProjectID links it to the owning project. The zero
+// value is not a valid chapter: AddChapter requires a non-blank title and
+// content, and generates the id.
+type Chapter struct {
+	ID        string
+	ProjectID string
+	Title     string
+	Content   string
+	CreatedAt time.Time
+	UpdatedAt time.Time
+}
+
+// AddChapter persists ch under projectID, assigning an id and missing
+// timestamps. It requires a non-blank projectID, title and content; a blank id
+// is generated. The project row itself does not have to exist, so chapters may
+// be added before or after their parent project, but DeleteProject cascades
+// them away together.
+func (s *Store) AddChapter(ctx context.Context, projectID string, ch Chapter) error {
+	if strings.TrimSpace(projectID) == "" {
+		return ErrMissingProject
+	}
+	if strings.TrimSpace(ch.Title) == "" {
+		return ErrMissingTitle
+	}
+	if strings.TrimSpace(ch.Content) == "" {
+		return ErrMissingContent
+	}
+	now := time.Now().UTC()
+	if ch.ID == "" {
+		id, err := newID()
+		if err != nil {
+			return err
+		}
+		ch.ID = id
+	}
+	if ch.CreatedAt.IsZero() {
+		ch.CreatedAt = now
+	}
+	if ch.UpdatedAt.IsZero() {
+		ch.UpdatedAt = now
+	}
+	_, err := s.db.ExecContext(ctx, `INSERT INTO project_chapter
+		(id, project_id, title, content, created_at, updated_at)
+		VALUES (?, ?, ?, ?, ?, ?)`,
+		ch.ID, projectID, ch.Title, ch.Content,
+		formatTime(ch.CreatedAt), formatTime(ch.UpdatedAt))
+	if err != nil {
+		return fmt.Errorf("projectindex: insert chapter: %w", err)
+	}
+	s.log.Debug("chapter added", "project_id", projectID, "chapter_id", ch.ID)
+	return nil
+}
+
+// ListChapters returns every chapter for projectID in insertion order. It returns
+// an empty (non-nil) slice when there are none; a blank projectID returns
+// ErrMissingProject.
+func (s *Store) ListChapters(ctx context.Context, projectID string) ([]Chapter, error) {
+	if strings.TrimSpace(projectID) == "" {
+		return nil, ErrMissingProject
+	}
+	rows, err := s.db.QueryContext(ctx, `SELECT id, project_id, title, content,
+		created_at, updated_at FROM project_chapter
+		WHERE project_id = ? ORDER BY created_at ASC, id ASC`, projectID)
+	if err != nil {
+		return nil, fmt.Errorf("projectindex: list chapters: %w", err)
+	}
+	defer func() { _ = rows.Close() }()
+
+	chapters := make([]Chapter, 0)
+	for rows.Next() {
+		ch, err := scanChapter(rows)
+		if err != nil {
+			return nil, fmt.Errorf("projectindex: scan chapter: %w", err)
+		}
+		chapters = append(chapters, ch)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("projectindex: iterate chapters: %w", err)
+	}
+	return chapters, nil
+}
+
+// scanChapter reads one project_chapter row.
+func scanChapter(sc rowScanner) (Chapter, error) {
+	var (
+		ch                  Chapter
+		createdAt, updatedAt string
+	)
+	if err := sc.Scan(&ch.ID, &ch.ProjectID, &ch.Title, &ch.Content, &createdAt, &updatedAt); err != nil {
+		return Chapter{}, err
+	}
+	var err error
+	if ch.CreatedAt, err = parseTime(createdAt); err != nil {
+		return Chapter{}, err
+	}
+	if ch.UpdatedAt, err = parseTime(updatedAt); err != nil {
+		return Chapter{}, err
+	}
+	return ch, nil
+}
diff --git a/internal/runtime/memory/projectindex/project.go b/internal/runtime/memory/projectindex/project.go
new file mode 100644
index 0000000..0126df6
--- /dev/null
+++ b/internal/runtime/memory/projectindex/project.go
@@ -0,0 +1,162 @@
+package projectindex
+
+import (
+	"context"
+	"crypto/rand"
+	"database/sql"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+)
+
+// Project is a project index entry (design §3): the project marker that links
+// its chapters, together with the project's description. It is layer 1 (index
+// entry) and layer 2 (description) of the hierarchy. The zero value is not a
+// valid project: ID and Title are required.
+type Project struct {
+	ID        string
+	Title     string
+	Desc      string
+	CreatedAt time.Time
+	UpdatedAt time.Time
+}
+
+// PutProject upserts p by id. On an existing id it updates title, desc, and
+// updated_at while preserving the original created_at; on a new id it assigns
+// missing timestamps and inserts. A blank id or title returns ErrMissingProject
+// or ErrMissingTitle.
+func (s *Store) PutProject(ctx context.Context, p Project) error {
+	if strings.TrimSpace(p.ID) == "" {
+		return ErrMissingProject
+	}
+	if strings.TrimSpace(p.Title) == "" {
+		return ErrMissingTitle
+	}
+	now := time.Now().UTC()
+	if p.CreatedAt.IsZero() {
+		p.CreatedAt = now
+	}
+	if p.UpdatedAt.IsZero() {
+		p.UpdatedAt = now
+	}
+	_, err := s.db.ExecContext(ctx, `INSERT INTO project (id, title, desc, created_at, updated_at)
+		VALUES (?, ?, ?, ?, ?)
+		ON CONFLICT(id) DO UPDATE SET
+			title = excluded.title,
+			desc = excluded.desc,
+			updated_at = excluded.updated_at`,
+		p.ID, p.Title, p.Desc, formatTime(p.CreatedAt), formatTime(p.UpdatedAt))
+	if err != nil {
+		return fmt.Errorf("projectindex: upsert project: %w", err)
+	}
+	s.log.Debug("project upserted", "id", p.ID)
+	return nil
+}
+
+// GetProject returns the project with the given id. It returns ErrProjectNotFound
+// when the id is blank or does not exist.
+func (s *Store) GetProject(ctx context.Context, id string) (*Project, error) {
+	if strings.TrimSpace(id) == "" {
+		return nil, ErrProjectNotFound
+	}
+	row := s.db.QueryRowContext(ctx, `SELECT id, title, desc, created_at, updated_at
+		FROM project WHERE id = ?`, id)
+	p, err := scanProject(row)
+	if errors.Is(err, sql.ErrNoRows) {
+		return nil, ErrProjectNotFound
+	}
+	if err != nil {
+		return nil, fmt.Errorf("projectindex: get project: %w", err)
+	}
+	return p, nil
+}
+
+// DeleteProject removes the project by id and, as a cascade, every chapter that
+// references it. It returns ErrProjectNotFound when the id is blank or no
+// project row was removed. The project removal and the chapter removal happen in
+// one transaction so a failure never leaves an orphaned project or chapter.
+func (s *Store) DeleteProject(ctx context.Context, id string) error {
+	if strings.TrimSpace(id) == "" {
+		return ErrProjectNotFound
+	}
+	tx, err := s.db.BeginTx(ctx, nil)
+	if err != nil {
+		return fmt.Errorf("projectindex: begin delete tx: %w", err)
+	}
+	defer func() { _ = tx.Rollback() }()
+
+	res, err := tx.ExecContext(ctx, `DELETE FROM project_chapter WHERE project_id = ?`, id)
+	if err != nil {
+		return fmt.Errorf("projectindex: delete chapters: %w", err)
+	}
+	if _, err := res.RowsAffected(); err != nil {
+		return fmt.Errorf("projectindex: delete chapters row count: %w", err)
+	}
+
+	res, err = tx.ExecContext(ctx, `DELETE FROM project WHERE id = ?`, id)
+	if err != nil {
+		return fmt.Errorf("projectindex: delete project: %w", err)
+	}
+	n, err := res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("projectindex: delete row count: %w", err)
+	}
+	if n == 0 {
+		return ErrProjectNotFound
+	}
+	if err := tx.Commit(); err != nil {
+		return fmt.Errorf("projectindex: commit delete tx: %w", err)
+	}
+	s.log.Debug("project deleted", "id", id)
+	return nil
+}
+
+// scanProject reads one project row.
+func scanProject(sc rowScanner) (*Project, error) {
+	var (
+		p                   Project
+		createdAt, updatedAt string
+	)
+	if err := sc.Scan(&p.ID, &p.Title, &p.Desc, &createdAt, &updatedAt); err != nil {
+		return nil, err
+	}
+	var err error
+	if p.CreatedAt, err = parseTime(createdAt); err != nil {
+		return nil, err
+	}
+	if p.UpdatedAt, err = parseTime(updatedAt); err != nil {
+		return nil, err
+	}
+	return &p, nil
+}
+
+// newID returns a 128-bit cryptographically random id, hex-encoded. It avoids a
+// UUID dependency so the package stays self-contained. rand.Read fails only if
+// the platform entropy source is broken, so the error is propagated.
+func newID() (string, error) {
+	b := make([]byte, 16)
+	if _, err := rand.Read(b); err != nil {
+		return "", fmt.Errorf("projectindex: generate id: %w", err)
+	}
+	return hex.EncodeToString(b), nil
+}
+
+// formatTime encodes t in UTC as an RFC3339Nano string so ordering and
+// round-trips are stable across processes.
+func formatTime(t time.Time) string {
+	return t.UTC().Format(time.RFC3339Nano)
+}
+
+// parseTime decodes a stored RFC3339 string; blank input yields a zero time.
+func parseTime(s string) (time.Time, error) {
+	if strings.TrimSpace(s) == "" {
+		return time.Time{}, nil
+	}
+	t, err := time.Parse(time.RFC3339Nano, s)
+	if err != nil {
+		return time.Time{}, fmt.Errorf("projectindex: parse time %q: %w", s, err)
+	}
+	return t, nil
+}
diff --git a/internal/runtime/memory/projectindex/projectindex_test.go b/internal/runtime/memory/projectindex/projectindex_test.go
new file mode 100644
index 0000000..4f997f2
--- /dev/null
+++ b/internal/runtime/memory/projectindex/projectindex_test.go
@@ -0,0 +1,498 @@
+package projectindex
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"git.yeij.top/AskaEth/Cyrene_v2/internal/db"
+)
+
+// testLogger returns a logger that writes nowhere, keeping test output clean.
+func testLogger() *slog.Logger {
+	return slog.New(slog.NewTextHandler(io.Discard, nil))
+}
+
+// newTestStore opens an isolated in-memory SQLite DB and builds a Store over it.
+// Each call owns a private DB, so parallel tests are isolated by construction.
+func newTestStore(t *testing.T) *Store {
+	t.Helper()
+	ctx := context.Background()
+	d, err := db.Open(ctx, ":memory:", testLogger())
+	if err != nil {
+		t.Fatalf("open in-memory db: %v", err)
+	}
+	t.Cleanup(func() { _ = d.Close() })
+
+	s, err := New(ctx, d, testLogger())
+	if err != nil {
+		t.Fatalf("New: %v", err)
+	}
+	return s
+}
+
+// baseProject is a minimal valid project for tests.
+func baseProject() Project {
+	return Project{ID: "proj.1", Title: "Cyrene", Desc: "陪伴型 AI 助手"}
+}
+
+func TestNew_NilDB_异常(t *testing.T) {
+	t.Parallel()
+	if _, err := New(context.Background(), nil, testLogger()); !errors.Is(err, ErrNilDB) {
+		t.Fatalf("New(nil db) err = %v, want ErrNilDB", err)
+	}
+}
+
+func TestNew_Schema_正常(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	d, err := db.Open(ctx, ":memory:", testLogger())
+	if err != nil {
+		t.Fatalf("open in-memory db: %v", err)
+	}
+	t.Cleanup(func() { _ = d.Close() })
+	if _, err := New(ctx, d, testLogger()); err != nil {
+		t.Fatalf("New: %v", err)
+	}
+
+	for _, tbl := range []string{"project", "project_chapter"} {
+		var n int
+		if err := d.QueryRowContext(ctx,
+			"SELECT count(*) FROM sqlite_master WHERE type='table' AND name=?", tbl).Scan(&n); err != nil {
+			t.Fatalf("check table %s: %v", tbl, err)
+		}
+		if n != 1 {
+			t.Fatalf("table %q missing", tbl)
+		}
+	}
+	var n int
+	if err := d.QueryRowContext(ctx,
+		"SELECT count(*) FROM sqlite_master WHERE type='index' AND name='idx_project_chapter_project_id'").Scan(&n); err != nil {
+		t.Fatalf("check index: %v", err)
+	}
+	if n != 1 {
+		t.Fatalf("index idx_project_chapter_project_id missing")
+	}
+}
+
+func TestStore_PutGetProject_Upsert(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	oldTime := time.Date(2020, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := s.PutProject(ctx, Project{ID: "proj.1", Title: "T1", Desc: "D1", CreatedAt: oldTime}); err != nil {
+		t.Fatalf("PutProject insert: %v", err)
+	}
+	// Upsert preserves the original created_at while updating title/desc.
+	if err := s.PutProject(ctx, Project{ID: "proj.1", Title: "T2", Desc: "D2", CreatedAt: oldTime.Add(time.Hour)}); err != nil {
+		t.Fatalf("PutProject update: %v", err)
+	}
+	got, err := s.GetProject(ctx, "proj.1")
+	if err != nil {
+		t.Fatalf("GetProject: %v", err)
+	}
+	if got.Title != "T2" {
+		t.Errorf("title = %q, want T2", got.Title)
+	}
+	if got.Desc != "D2" {
+		t.Errorf("desc = %q, want D2", got.Desc)
+	}
+	if !got.CreatedAt.Equal(oldTime) {
+		t.Errorf("created_at = %v, want preserved %v", got.CreatedAt, oldTime)
+	}
+	if got.UpdatedAt.IsZero() {
+		t.Errorf("updated_at not set")
+	}
+}
+
+func TestStore_PutProject_Validation_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	tests := []struct {
+		name    string
+		p       Project
+		wantErr error
+	}{
+		{name: "blank id", p: Project{Title: "T"}, wantErr: ErrMissingProject},
+		{name: "blank title", p: Project{ID: "p"}, wantErr: ErrMissingTitle},
+	}
+	for _, tt := range tests {
+		tt := tt
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			if err := s.PutProject(ctx, tt.p); !errors.Is(err, tt.wantErr) {
+				t.Fatalf("PutProject err = %v, want wrap of %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+func TestStore_GetProject_NotFound_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	for _, id := range []string{"", "does-not-exist"} {
+		if _, err := s.GetProject(ctx, id); !errors.Is(err, ErrProjectNotFound) {
+			t.Fatalf("GetProject(%q) err = %v, want ErrProjectNotFound", id, err)
+		}
+	}
+}
+
+func TestStore_AddListChapters_正常(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "概述", Content: "第一行\n第二行"}); err != nil {
+		t.Fatalf("AddChapter 1: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "实现", Content: "另外一个章节"}); err != nil {
+		t.Fatalf("AddChapter 2: %v", err)
+	}
+
+	list, err := s.ListChapters(ctx, "proj.1")
+	if err != nil {
+		t.Fatalf("ListChapters: %v", err)
+	}
+	if len(list) != 2 {
+		t.Fatalf("ListChapters = %d chapters, want 2", len(list))
+	}
+	first := list[0]
+	if first.Title != "概述" || first.Content != "第一行\n第二行" {
+		t.Errorf("first chapter = %+v, want 概述/第一行", first)
+	}
+	if first.ProjectID != "proj.1" {
+		t.Errorf("ProjectID = %q, want proj.1", first.ProjectID)
+	}
+	if first.ID == "" {
+		t.Errorf("AddChapter did not assign an id")
+	}
+	if first.CreatedAt.IsZero() || first.UpdatedAt.IsZero() {
+		t.Errorf("AddChapter did not assign timestamps")
+	}
+
+	// A different project must not see these chapters.
+	other, err := s.ListChapters(ctx, "proj.other")
+	if err != nil {
+		t.Fatalf("ListChapters(proj.other): %v", err)
+	}
+	if len(other) != 0 {
+		t.Fatalf("ListChapters(proj.other) = %d chapters, want 0", len(other))
+	}
+}
+
+func TestStore_AddChapter_Validation_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	tests := []struct {
+		name      string
+		projectID string
+		ch        Chapter
+		wantErr   error
+	}{
+		{name: "blank project", ch: Chapter{Title: "T", Content: "C"}, wantErr: ErrMissingProject},
+		{name: "blank title", projectID: "p", ch: Chapter{Content: "C"}, wantErr: ErrMissingTitle},
+		{name: "blank content", projectID: "p", ch: Chapter{Title: "T"}, wantErr: ErrMissingContent},
+	}
+	for _, tt := range tests {
+		tt := tt
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			if err := s.AddChapter(ctx, tt.projectID, tt.ch); !errors.Is(err, tt.wantErr) {
+				t.Fatalf("AddChapter err = %v, want wrap of %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+func TestStore_ListChapters_MissingProject_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	if _, err := s.ListChapters(context.Background(), ""); !errors.Is(err, ErrMissingProject) {
+		t.Fatalf("ListChapters(\"\") err = %v, want ErrMissingProject", err)
+	}
+}
+
+func TestStore_Search_HitProject(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, Project{ID: "proj.1", Title: "Cyrene 重构", Desc: "陪伴型 AI 助手"}); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "实现", Content: "纯文本内容"}); err != nil {
+		t.Fatalf("AddChapter: %v", err)
+	}
+
+	res, err := s.Search(ctx, "proj.1", "陪伴型")
+	if err != nil {
+		t.Fatalf("Search: %v", err)
+	}
+	if res.Project == nil {
+		t.Fatalf("Search result.Project nil, want project hit (desc layer)")
+	}
+	if res.Project.ID != "proj.1" || res.Project.Desc != "陪伴型 AI 助手" {
+		t.Errorf("Project = %+v, want proj.1/陪伴型 AI 助手", res.Project)
+	}
+	if len(res.Chapters) != 0 {
+		t.Fatalf("Search result.Chapters = %d, want 0 (chapter content did not match)", len(res.Chapters))
+	}
+}
+
+func TestStore_Search_HitChapter_分层(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	content := "第一行\n核心算法在第二行\n第三行\n第四行\n第五行"
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "实现", Content: content}); err != nil {
+		t.Fatalf("AddChapter: %v", err)
+	}
+
+	res, err := s.Search(ctx, "proj.1", "算法")
+	if err != nil {
+		t.Fatalf("Search: %v", err)
+	}
+	if res.Project != nil {
+		t.Fatalf("Search result.Project = %v, want nil (project layer did not match)", res.Project)
+	}
+	if len(res.Chapters) != 1 {
+		t.Fatalf("Search result.Chapters = %d, want 1", len(res.Chapters))
+	}
+	hit := res.Chapters[0]
+	if hit.Chapter.Title != "实现" || hit.Chapter.Content != content {
+		t.Errorf("Chapter = %+v, want 实现", hit.Chapter)
+	}
+	if len(hit.Matches) != 1 {
+		t.Fatalf("Matches = %d, want 1", len(hit.Matches))
+	}
+	m := hit.Matches[0]
+	if m.LineNumber != 2 {
+		t.Errorf("LineNumber = %d, want 2", m.LineNumber)
+	}
+	if m.Text != "核心算法在第二行" {
+		t.Errorf("Text = %q, want 核心算法在第二行", m.Text)
+	}
+	wantContext := "第一行\n核心算法在第二行\n第三行\n第四行"
+	if m.Context != wantContext {
+		t.Errorf("Context = %q, want %q", m.Context, wantContext)
+	}
+}
+
+func TestStore_Search_NoHit(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "实现", Content: "纯文本内容"}); err != nil {
+		t.Fatalf("AddChapter: %v", err)
+	}
+
+	res, err := s.Search(ctx, "proj.1", "不存在的关键词")
+	if err != nil {
+		t.Fatalf("Search: %v", err)
+	}
+	if res.Project != nil {
+		t.Errorf("Project = %v, want nil", res.Project)
+	}
+	if len(res.Chapters) != 0 {
+		t.Errorf("Chapters = %d, want 0", len(res.Chapters))
+	}
+}
+
+func TestStore_Search_ProjectNotFound_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	if _, err := s.Search(context.Background(), "missing", "kw"); !errors.Is(err, ErrProjectNotFound) {
+		t.Fatalf("Search(missing) err = %v, want ErrProjectNotFound", err)
+	}
+}
+
+func TestStore_Search_BlankKeyword_异常(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	if _, err := s.Search(ctx, "proj.1", "   "); !errors.Is(err, ErrMissingKeyword) {
+		t.Fatalf("Search(blank) err = %v, want ErrMissingKeyword", err)
+	}
+}
+
+func TestStore_DeleteProject_Cascade(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "c1", Content: "x"}); err != nil {
+		t.Fatalf("AddChapter c1: %v", err)
+	}
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "c2", Content: "y"}); err != nil {
+		t.Fatalf("AddChapter c2: %v", err)
+	}
+
+	if err := s.DeleteProject(ctx, "proj.1"); err != nil {
+		t.Fatalf("DeleteProject: %v", err)
+	}
+	if _, err := s.GetProject(ctx, "proj.1"); !errors.Is(err, ErrProjectNotFound) {
+		t.Fatalf("GetProject after delete err = %v, want ErrProjectNotFound", err)
+	}
+	ch, err := s.ListChapters(ctx, "proj.1")
+	if err != nil {
+		t.Fatalf("ListChapters after delete: %v", err)
+	}
+	if len(ch) != 0 {
+		t.Fatalf("chapters not cascaded, got %d", len(ch))
+	}
+}
+
+func TestStore_DeleteProject_NotFound_异常(t *testing.T) {
+	t.Parallel()
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	for _, id := range []string{"", "missing"} {
+		if err := s.DeleteProject(ctx, id); !errors.Is(err, ErrProjectNotFound) {
+			t.Fatalf("DeleteProject(%q) err = %v, want ErrProjectNotFound", id, err)
+		}
+	}
+}
+
+func TestStore_Search_SQLInjectionSafe(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	maliciousID := "proj.1; DROP TABLE project; --"
+	if err := s.PutProject(ctx, Project{ID: maliciousID, Title: "恶意', `x`", Desc: "描述"}); err != nil {
+		t.Fatalf("PutProject malicious: %v", err)
+	}
+	if err := s.AddChapter(ctx, maliciousID, Chapter{Title: "c'--", Content: "含有 keyword 的行\n普通行"}); err != nil {
+		t.Fatalf("AddChapter malicious: %v", err)
+	}
+
+	// A keyword full of SQL metacharacters must not alter behavior or data.
+	res, err := s.Search(ctx, maliciousID, "' OR 1=1 --")
+	if err != nil {
+		t.Fatalf("Search malicious keyword: %v", err)
+	}
+	if res.Project != nil {
+		t.Errorf("Project = %v, want nil for non-matching keyword", res.Project)
+	}
+	if len(res.Chapters) != 0 {
+		t.Errorf("Chapters = %d, want 0 for non-matching keyword", len(res.Chapters))
+	}
+	if _, err := s.Search(ctx, maliciousID, "zzz_never"); err != nil {
+		t.Fatalf("Search benign keyword on malicious id: %v", err)
+	}
+
+	// The malicious id must round-trip intact (proves parameterized lookup).
+	got, err := s.GetProject(ctx, maliciousID)
+	if err != nil {
+		t.Fatalf("GetProject malicious: %v", err)
+	}
+	if got.ID != maliciousID {
+		t.Errorf("ID = %q, want %q", got.ID, maliciousID)
+	}
+	// And the malicious project must still be queryable, not dropped.
+	if _, err := s.GetProject(ctx, "project"); !errors.Is(err, ErrProjectNotFound) {
+		t.Fatalf("project table intact check: err = %v, want ErrProjectNotFound", err)
+	}
+}
+
+func TestStore_Concurrent_AddChapter_并发(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+
+	const n = 30
+	var wg sync.WaitGroup
+	errs := make(chan error, n)
+	for i := 0; i < n; i++ {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			errs <- s.AddChapter(ctx, "proj.1", Chapter{Title: fmt.Sprintf("ch%d", i), Content: "并发内容"})
+		}(i)
+	}
+	wg.Wait()
+	close(errs)
+
+	for err := range errs {
+		if err != nil {
+			t.Fatalf("concurrent AddChapter error: %v", err)
+		}
+	}
+	ch, err := s.ListChapters(ctx, "proj.1")
+	if err != nil {
+		t.Fatalf("ListChapters: %v", err)
+	}
+	if len(ch) != n {
+		t.Fatalf("ListChapters = %d chapters, want %d", len(ch), n)
+	}
+}
+
+func TestStore_Search_MatchInMultipleLines(t *testing.T) {
+	t.Parallel()
+	ctx := context.Background()
+	s := newTestStore(t)
+
+	if err := s.PutProject(ctx, baseProject()); err != nil {
+		t.Fatalf("PutProject: %v", err)
+	}
+	content := "关于 AI\nAI 是一个话题\n无关行\n结尾 AI 又出现"
+	if err := s.AddChapter(ctx, "proj.1", Chapter{Title: "概述", Content: content}); err != nil {
+		t.Fatalf("AddChapter: %v", err)
+	}
+
+	res, err := s.Search(ctx, "proj.1", "ai")
+	if err != nil {
+		t.Fatalf("Search: %v", err)
+	}
+	if len(res.Chapters) != 1 {
+		t.Fatalf("Chapters = %d, want 1", len(res.Chapters))
+	}
+	hit := res.Chapters[0]
+	if len(hit.Matches) != 3 {
+		t.Fatalf("Matches = %d, want 3", len(hit.Matches))
+	}
+	if hit.Matches[0].LineNumber != 1 || hit.Matches[1].LineNumber != 2 || hit.Matches[2].LineNumber != 4 {
+		t.Errorf("line numbers = [%d %d %d], want [1 2 4]",
+			hit.Matches[0].LineNumber, hit.Matches[1].LineNumber, hit.Matches[2].LineNumber)
+	}
+	// The keyword was lowercase "ai" but the content holds uppercase "AI",
+	// so the match is case-insensitive while Text preserves the original casing.
+	if !strings.Contains(hit.Matches[0].Text, "AI") {
+		t.Errorf("expected case-insensitive match, Text = %q", hit.Matches[0].Text)
+	}
+}
diff --git a/internal/runtime/memory/projectindex/search.go b/internal/runtime/memory/projectindex/search.go
new file mode 100644
index 0000000..8bcdd29
--- /dev/null
+++ b/internal/runtime/memory/projectindex/search.go
@@ -0,0 +1,108 @@
+package projectindex
+
+import (
+	"context"
+	"strings"
+)
+
+// contextRadius is the number of lines shown on each side of a matched line in
+// LineMatch.Context.
+const contextRadius = 2
+
+// LineMatch is a single matched line inside a chapter's content plus its
+// surrounding context. LineNumber is 1-based within the chapter's Content.
+type LineMatch struct {
+	LineNumber int
+	Text       string
+	Context    string
+}
+
+// ChapterHit is a chapter whose content matched a keyword, with the located
+// lines (and their context) that matched.
+type ChapterHit struct {
+	Chapter *Chapter
+	Matches []LineMatch
+}
+
+// SearchResult is the hierarchical result of a Search. Project carries the
+// matched project index entry + description (layer 1/2), set only when the
+// keyword hits the project title or desc; Chapters carries the matched chapters
+// (layer 3) with located lines. When nothing matches, Project is nil and
+// Chapters is empty, with no error.
+type SearchResult struct {
+	Project  *Project
+	Chapters []ChapterHit
+}
+
+// Search looks up keyword within a single project, walking the hierarchy
+// 项目索引 → 项目描述 → 章节内容. It first loads the project (ErrProjectNotFound
+// when it does not exist), sets SearchResult.Project when the keyword hits the
+// project title or description, then scans every chapter for content hits,
+// producing located lines plus their context. A blank keyword returns
+// ErrMissingKeyword; a keyword that hits nothing returns a non-nil SearchResult
+// with Project and Chapters unset.
+func (s *Store) Search(ctx context.Context, projectID, keyword string) (*SearchResult, error) {
+	kw := strings.ToLower(strings.TrimSpace(keyword))
+	if kw == "" {
+		return nil, ErrMissingKeyword
+	}
+	proj, err := s.GetProject(ctx, projectID)
+	if err != nil {
+		return nil, err
+	}
+	result := &SearchResult{}
+	if containsFold(proj.Title, kw) || containsFold(proj.Desc, kw) {
+		result.Project = proj
+	}
+	chapters, err := s.ListChapters(ctx, projectID)
+	if err != nil {
+		return nil, err
+	}
+	for i := range chapters {
+		matches := findLineMatches(chapters[i].Content, kw)
+		if len(matches) == 0 {
+			continue
+		}
+		ch := chapters[i]
+		result.Chapters = append(result.Chapters, ChapterHit{Chapter: &ch, Matches: matches})
+	}
+	return result, nil
+}
+
+// containsFold reports whether s (case-insensitively) contains sub, which must
+// already be lowercased by the caller.
+func containsFold(s, sub string) bool {
+	return strings.Contains(strings.ToLower(s), sub)
+}
+
+// findLineMatches returns one LineMatch per content line that (case-insensitively)
+// contains keyword. Context is the surrounding contextRadius lines on each side,
+// joined by newlines and including the matched line.
+func findLineMatches(content, keyword string) []LineMatch {
+	lines := strings.Split(content, "\n")
+	matches := make([]LineMatch, 0)
+	for i, line := range lines {
+		if strings.Contains(strings.ToLower(line), keyword) {
+			matches = append(matches, LineMatch{
+				LineNumber: i + 1,
+				Text:       line,
+				Context:    buildContext(lines, i),
+			})
+		}
+	}
+	return matches
+}
+
+// buildContext returns lines[start..end] around idx (inclusive) joined by \n,
+// clamped to the slice bounds so the context never overflows.
+func buildContext(lines []string, idx int) string {
+	start := idx - contextRadius
+	if start < 0 {
+		start = 0
+	}
+	end := idx + contextRadius
+	if end >= len(lines) {
+		end = len(lines) - 1
+	}
+	return strings.Join(lines[start:end+1], "\n")
+}
diff --git a/internal/runtime/memory/projectindex/store.go b/internal/runtime/memory/projectindex/store.go
new file mode 100644
index 0000000..46ab836
--- /dev/null
+++ b/internal/runtime/memory/projectindex/store.go
@@ -0,0 +1,132 @@
+// Package projectindex implements the project-index hierarchy from
+// DevDocs/design/03-记忆.md §3 ("项目记忆"作用域): a project is indexed as an
+// entry plus description, and its content is split into chapters that can be
+// searched line by line. The three-layer depth is 项目索引 → 项目描述 → 章节内容.
+//
+// The two tables model that depth: the project row is the index entry (layer 1)
+// carrying its description (layer 2), and project_chapter holds the chapter
+// content (layer 3). Project memory is readable and writable by every persona
+// (design §3 "人人可读可写", "项目索引各角色可查"), so the store carries no
+// persona/role column and never filters by identity here.
+//
+// This package is deliberately separate from the vector index
+// (internal/runtime/memory/vector): it does text-based hierarchical indexing of
+// project entries, descriptions, and chapters. Semantic/vector search belongs to
+// the vector package.
+//
+// The store owns its schema: New creates the project and project_chapter tables
+// and the chapter index idempotently. All SQL is parameterized (?), column lists
+// are explicit (no SELECT *), reads take a context.Cancel, and every error is
+// wrapped with %w so callers can use errors.Is against the package sentinels.
+package projectindex
+
+import (
+	"context"
+	"database/sql"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+
+	"git.yeij.top/AskaEth/Cyrene_v2/internal/logger"
+)
+
+// Sentinels distinguishing projectindex failures. Each is wrapped with %w at the
+// call site so callers can use errors.Is.
+var (
+	// ErrProjectNotFound is returned when a project id is blank or does not exist.
+	ErrProjectNotFound = errors.New("projectindex: project not found")
+
+	// ErrChapterNotFound is returned when a required chapter reference is invalid.
+	ErrChapterNotFound = errors.New("projectindex: chapter not found")
+
+	// ErrNilDB is returned when New receives a nil *sql.DB.
+	ErrNilDB = errors.New("projectindex: nil db")
+
+	// ErrMissingProject is returned when a required project id is blank.
+	ErrMissingProject = errors.New("projectindex: project_id is required")
+
+	// ErrMissingTitle is returned when a project or chapter title is blank.
+	ErrMissingTitle = errors.New("projectindex: title is required")
+
+	// ErrMissingContent is returned when a chapter has no content.
+	ErrMissingContent = errors.New("projectindex: content is required")
+
+	// ErrMissingKeyword is returned when Search receives a blank keyword.
+	ErrMissingKeyword = errors.New("projectindex: keyword is required")
+)
+
+// Store is a project-index store backed by a *sql.DB.
+//
+// It owns its schema: New creates the project / project_chapter tables and the
+// chapter index idempotently. Concurrency is delegated to *sql.DB, which is safe
+// for concurrent use by contract, so Store holds no further lock. Every query is
+// parameterized and reads take a context.Cancel.
+type Store struct {
+	db  *sql.DB
+	log *slog.Logger
+}
+
+// rowScanner is satisfied by both *sql.Row and *sql.Rows so a single scanProject
+// or scanChapter helper serves both the Get and List paths.
+type rowScanner interface {
+	Scan(dest ...any) error
+}
+
+// New builds a Store on db, ensuring the schema exists.
+//
+// db must be a live connection (the migration framework is owned by
+// internal/db). A nil db returns ErrNilDB. log may be nil, in which case a
+// discard logger is used; the active logger is tagged with source
+// "memory/projectindex" so every record is traceable. ctx controls schema
+// creation and must be honored by the caller.
+func New(ctx context.Context, db *sql.DB, log *slog.Logger) (*Store, error) {
+	if db == nil {
+		return nil, ErrNilDB
+	}
+	log = logger.WithSource(discardLogger(log), "memory/projectindex")
+	for _, ddl := range []string{
+		createProjectTable,
+		createProjectChapterTable,
+		createProjectChapterIndex,
+	} {
+		if _, err := db.ExecContext(ctx, ddl); err != nil {
+			return nil, fmt.Errorf("projectindex: apply schema: %w", err)
+		}
+	}
+	return &Store{db: db, log: log}, nil
+}
+
+// discardLogger returns log if non-nil, otherwise a logger that writes nowhere.
+func discardLogger(log *slog.Logger) *slog.Logger {
+	if log != nil {
+		return log
+	}
+	return slog.New(slog.NewTextHandler(io.Discard, nil))
+}
+
+// Schema DDL, applied idempotently by New. The two tables model the three-layer
+// hierarchy: the project row is the index entry (layer 1) plus its description
+// (layer 2), and project_chapter holds the chapter content (layer 3).
+const createProjectTable = `
+CREATE TABLE IF NOT EXISTS project (
+    id         TEXT PRIMARY KEY,
+    title      TEXT NOT NULL,
+    desc       TEXT NOT NULL DEFAULT '',
+    created_at TEXT NOT NULL,
+    updated_at TEXT NOT NULL
+)`
+
+const createProjectChapterTable = `
+CREATE TABLE IF NOT EXISTS project_chapter (
+    id         TEXT PRIMARY KEY,
+    project_id TEXT NOT NULL,
+    title      TEXT NOT NULL,
+    content    TEXT NOT NULL,
+    created_at TEXT NOT NULL,
+    updated_at TEXT NOT NULL
+)`
+
+const createProjectChapterIndex = `
+CREATE INDEX IF NOT EXISTS idx_project_chapter_project_id
+    ON project_chapter (project_id)`