diff --git a/js/packages/cli/src/api/routes.ts b/js/packages/cli/src/api/routes.ts
index 8dc2efef..78ca9e4f 100644
--- a/js/packages/cli/src/api/routes.ts
+++ b/js/packages/cli/src/api/routes.ts
@@ -8,7 +8,7 @@
  * is the part nobody notices — which secret a given entry produces.
  */
 
-import { readFile, stat } from "node:fs/promises";
+import { readFile, mkdir } from "node:fs/promises";
 import { existsSync } from "node:fs";
 import { basename, dirname, join, resolve as resolvePath } from "node:path";
 import {
@@ -38,6 +38,11 @@ import {
   passwordPolicyFromRecord,
   removeLink,
   restoreEntry,
+  buildSemanticRecords,
+  searchSemanticRecords,
+  semanticManifest,
+  semanticStatus,
+  type SemanticRecord,
   totpCodeAt,
   Bip85,
   type Entry,
@@ -64,7 +69,6 @@ export const UNPORTED_PREFIXES: Array<{ prefix: string; feature: string }> = [
   { prefix: "/api/v1/high-risk", feature: "high-risk partitions" },
   { prefix: "/api/v1/agent/job-profiles", feature: "agent job profiles" },
   { prefix: "/api/v1/agent/recovery", feature: "agent recovery split" },
-  { prefix: "/api/v1/semantic", feature: "the semantic (vector search) index" },
 ];
 
 export interface ApiContext {
@@ -812,6 +816,84 @@ export function registerRoutes(server: ApiServer, ctx: ApiContext): void {
     };
   });
 
+  // --------------------------------------------------------------- semantic
+
+  const semanticDir = (): string => join(profileDir(ctx), "semantic_index");
+  const semanticRecordsPath = (): string => join(semanticDir(), "records.json");
+  const semanticManifestPath = (): string => join(semanticDir(), "manifest.json");
+
+  async function readSemanticRecords(): Promise<SemanticRecord[]> {
+    if (!existsSync(semanticRecordsPath())) return [];
+    try {
+      const data = JSON.parse(await readFile(semanticRecordsPath(), "utf8"));
+      return Array.isArray(data) ? (data as SemanticRecord[]) : [];
+    } catch {
+      return [];
+    }
+  }
+
+  async function buildSemantic(): Promise<number> {
+    const vault = await readVault(ctx);
+    const entries = Object.entries(vault.index.entries).map(([id, e]) => ({
+      ...(e as object),
+      id: Number(id),
+    })) as Array<Record<string, unknown>>;
+    const records = buildSemanticRecords(entries);
+    await mkdir(semanticDir(), { recursive: true });
+    // The records file is plaintext, so it holds metadata only and is written
+    // through atomicWrite for a genuine 0600 inode.
+    await atomicWrite(semanticRecordsPath(), new TextEncoder().encode(JSON.stringify(records, null, 2)));
+    await atomicWrite(
+      semanticManifestPath(),
+      new TextEncoder().encode(
+        JSON.stringify(
+          semanticManifest({
+            enabled: true,
+            built: true,
+            recordCount: records.length,
+            updatedAt: ctx.now() / 1000,
+          }),
+          null,
+          2,
+        ),
+      ),
+    );
+    return records.length;
+  }
+
+  server.route("GET", "/api/v1/semantic/status", async () => {
+    let manifest: Record<string, unknown> = {};
+    if (existsSync(semanticManifestPath())) {
+      try {
+        manifest = JSON.parse(await readFile(semanticManifestPath(), "utf8"));
+      } catch {
+        // Corrupt manifest reports as not-built; the fix is a rebuild.
+      }
+    }
+    return { json: semanticStatus(manifest, (await readSemanticRecords()).length) };
+  });
+
+  for (const path of ["/api/v1/semantic/build", "/api/v1/semantic/rebuild"]) {
+    server.route("POST", path, async () => {
+      // Build and rebuild are the same operation here: the records file is
+      // rewritten wholesale either way, so there is no stale state to clear.
+      return { json: { status: "ok", records: await buildSemantic() } };
+    });
+  }
+
+  server.route("POST", "/api/v1/semantic/search", async (req) => {
+    const body = bodyObject(req);
+    const records = await readSemanticRecords();
+    if (records.length === 0) {
+      throw new HttpError(409, "no semantic index; POST /api/v1/semantic/build first");
+    }
+    const hits = searchSemanticRecords(records, str(body["query"], "query"), {
+      k: optInt(body["k"], "k") ?? 10,
+      ...(typeof body["kind"] === "string" && { kind: body["kind"] }),
+    });
+    return { json: { results: hits.map((h) => ({ ...h, ref: refFor(String(h.entry_id)) })) } };
+  });
+
   server.route("POST", "/api/v1/shutdown", async () => {
     ctx.requestShutdown();
     return { json: { status: "shutting down" } };
diff --git a/js/packages/cli/src/capabilities.ts b/js/packages/cli/src/capabilities.ts
index ffc5d73e..e9618947 100644
--- a/js/packages/cli/src/capabilities.ts
+++ b/js/packages/cli/src/capabilities.ts
@@ -99,7 +99,6 @@ export function capabilities(): Record<string, unknown> {
       "pgp RSA keys (ed25519 is supported; RSA generation is not reproducible)",
       "index0/atlas",
       "approval gates and high-risk partitions",
-      "semantic (vector search) command group",
       "Python's v2/v3 TUIs (interactive mode follows the legacy v1 menus)",
     ],
   };
diff --git a/js/packages/cli/src/program.ts b/js/packages/cli/src/program.ts
index b4afa853..cc5c8c08 100644
--- a/js/packages/cli/src/program.ts
+++ b/js/packages/cli/src/program.ts
@@ -9,7 +9,7 @@
 
 import { Command } from "commander";
 import process from "node:process";
-import { entryMetadata, parseRef, resolveEntry } from "./refs.js";
+import { entryMetadata, parseRef, resolveEntry, refFor } from "./refs.js";
 import { materializeSecret } from "./secrets.js";
 import { clipboardSink, execSink, parseCommandSpec, stdinSink } from "./sinks.js";
 import { capabilities } from "./capabilities.js";
@@ -26,6 +26,11 @@ import {
   parseBackupWrapper,
   exportBackup,
   findDerivationCollisions,
+  buildSemanticRecords,
+  searchSemanticRecords,
+  semanticManifest,
+  semanticStatus,
+  type SemanticRecord,
   generateFingerprint,
   assertValidMnemonic,
   generateMnemonic,
@@ -1669,6 +1674,114 @@ export function buildProgram(io: ProgramIo = defaultIo): Command {
       io.out(JSON.stringify({ status: "shutting down", host: o.host, port }));
     });
 
+  const semantic = program
+    .command("semantic")
+    .description("local retrieval index over entry METADATA (never secrets)");
+
+  /**
+   * Records live in a plaintext file beside the vault, so the index holds
+   * only what is safe to leave unencrypted: labels, notes, tags, usernames,
+   * URLs, key names. Never a value. See buildSemanticRecords.
+   */
+  async function semanticPaths(opts: GlobalOpts) {
+    const app = new AppDir(resolveAppDir(opts.appDir));
+    const fp = await currentFingerprint(app, opts);
+    const dir = join(app.profileDir(fp), "semantic_index");
+    return { dir, records: join(dir, "records.json"), manifest: join(dir, "manifest.json") };
+  }
+
+  async function readRecords(path: string): Promise<SemanticRecord[]> {
+    if (!existsSync(path)) return [];
+    try {
+      const data = JSON.parse(await readFile(path, "utf8"));
+      return Array.isArray(data) ? (data as SemanticRecord[]) : [];
+    } catch {
+      return [];
+    }
+  }
+
+  semantic
+    .command("build")
+    .description("(re)build the index from the current vault")
+    .action(async () => {
+      const opts = program.opts() as GlobalOpts;
+      const vault = await openFromOptions(opts);
+      const paths = await semanticPaths(opts);
+      const entries = Object.entries(vault.index.entries).map(([id, e]) => ({
+        ...(e as object),
+        id: Number(id),
+      })) as Array<Record<string, unknown>>;
+      const records = buildSemanticRecords(entries);
+      await mkdir(paths.dir, { recursive: true });
+      // atomicWrite gives a fresh 0600 inode. Python created these at the
+      // process umask, which on a default install is 0664 — group and world
+      // readable, for a file sitting next to an encrypted vault.
+      await atomicWrite(
+        paths.records,
+        utf8(JSON.stringify(records, null, 2)),
+      );
+      await atomicWrite(
+        paths.manifest,
+        utf8(
+          JSON.stringify(
+            semanticManifest({
+              enabled: true,
+              built: true,
+              recordCount: records.length,
+              updatedAt: Date.now() / 1000,
+            }),
+            null,
+            2,
+          ),
+        ),
+      );
+      io.out(JSON.stringify({ built: records.length, index: paths.dir }));
+    });
+
+  semantic
+    .command("status")
+    .description("show whether the index exists and how current it is")
+    .action(async () => {
+      const opts = program.opts() as GlobalOpts;
+      const paths = await semanticPaths(opts);
+      let manifest: Record<string, unknown> = {};
+      if (existsSync(paths.manifest)) {
+        try {
+          manifest = JSON.parse(await readFile(paths.manifest, "utf8"));
+        } catch {
+          // A corrupt manifest reports as "not built" rather than failing:
+          // the fix is to rebuild, which the status is meant to prompt.
+        }
+      }
+      const records = await readRecords(paths.records);
+      io.out(JSON.stringify(semanticStatus(manifest, records.length), null, 2));
+    });
+
+  semantic
+    .command("search <query>")
+    .description("rank entries by metadata overlap (returns references, no secrets)")
+    .option("-k, --limit <n>", "maximum hits", "10")
+    .option("--kind <kind>", "restrict to one entry kind")
+    .action(async (query: string, o: { limit: string; kind?: string }) => {
+      const opts = program.opts() as GlobalOpts;
+      const paths = await semanticPaths(opts);
+      const records = await readRecords(paths.records);
+      if (records.length === 0) {
+        throw new Error("no semantic index; run 'semantic build' first");
+      }
+      const hits = searchSemanticRecords(records, query, {
+        k: parseIntOption(o.limit, "--limit", { min: 1 }),
+        ...(o.kind !== undefined && { kind: o.kind }),
+      });
+      io.out(
+        JSON.stringify(
+          hits.map((h) => ({ ...h, ref: refFor(String(h.entry_id)) })),
+          null,
+          2,
+        ),
+      );
+    });
+
   const util = program.command("util").description("utility commands");
 
   util
diff --git a/js/packages/cli/test/api.test.ts b/js/packages/cli/test/api.test.ts
index 25d07251..7e2a815a 100644
--- a/js/packages/cli/test/api.test.ts
+++ b/js/packages/cli/test/api.test.ts
@@ -10,7 +10,7 @@
  */
 
 import { afterAll, beforeAll, describe, expect, it } from "vitest";
-import { mkdtemp, mkdir, readFile, writeFile } from "node:fs/promises";
+import { mkdtemp, mkdir, readFile, writeFile, stat } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import process from "node:process";
@@ -430,13 +430,77 @@ describe("vault export and import", () => {
   });
 });
 
+describe("semantic index", () => {
+  it("builds, reports status, and ranks by metadata", async () => {
+    const built = await call("POST", "/api/v1/semantic/build");
+    expect(built.status).toBe(200);
+    expect(built.json.records).toBeGreaterThan(0);
+
+    const status = await call("GET", "/api/v1/semantic/status");
+    expect(status.json.built).toBe(true);
+    expect(status.json.model_id).toBe("seedpass-token-overlap-v2");
+
+    const hits = await call("POST", "/api/v1/semantic/search", { body: { query: "bank" } });
+    expect(hits.json.results.length).toBeGreaterThan(0);
+    expect(hits.json.results[0].ref).toMatch(/^sp:\/\/entry\//);
+  });
+
+  it("never writes a stored secret into the plaintext index", async () => {
+    await call("POST", "/api/v1/semantic/build");
+    // The vault holds a key_value entry whose value is this string. The
+    // index file sits unencrypted next to the vault, so the value must not
+    // reach it — nor a tokenized form of it.
+    const raw = await readFile(
+      join(app.profileDir(FINGERPRINT), "semantic_index", "records.json"),
+      "utf8",
+    );
+    expect(raw).not.toContain("kv-secret-value");
+    expect(raw.toLowerCase()).not.toContain("kv-secret-value".toLowerCase());
+    // But the entry is still findable by its name.
+    const hits = await call("POST", "/api/v1/semantic/search", { body: { query: "api-token" } });
+    expect(hits.json.results.map((r: any) => r.label)).toContain("api-token");
+  });
+
+  it("writes the index at 0600, not the process umask", async () => {
+    await call("POST", "/api/v1/semantic/build");
+    const path = join(app.profileDir(FINGERPRINT), "semantic_index", "records.json");
+    expect((await stat(path)).mode & 0o777).toBe(0o600);
+  });
+
+  it("says so rather than returning nothing when no index exists", async () => {
+    const fresh = new ApiServer({ host: "127.0.0.1", port: 0, token: TOKEN });
+    const other = await mkdtemp(join(tmpdir(), "seedpass-nosem-"));
+    const otherApp = new AppDir(other);
+    await otherApp.mutateFingerprints((d) => {
+      d.fingerprints.push(FINGERPRINT);
+      d.last_used = FINGERPRINT;
+    });
+    await mkdir(otherApp.profileDir(FINGERPRINT), { recursive: true });
+    await writeFile(
+      join(otherApp.profileDir(FINGERPRINT), INDEX_FILENAME),
+      await encryptV3(deriveIndexKeyBytes(MNEMONIC), utf8(JSON.stringify({ schema_version: 4, entries: {} }))),
+    );
+    registerRoutes(fresh, buildContext({ app: otherApp, fingerprint: FINGERPRINT, mnemonic: MNEMONIC }));
+    const bound = await fresh.listen();
+    try {
+      const res = await fetch(`http://${bound.host}:${bound.port}/api/v1/semantic/search`, {
+        method: "POST",
+        headers: { authorization: `Bearer ${TOKEN}`, "content-type": "application/json" },
+        body: JSON.stringify({ query: "anything" }),
+      });
+      expect(res.status).toBe(409);
+    } finally {
+      await fresh.close();
+    }
+  });
+});
+
 describe("the unported surface answers honestly", () => {
   it("returns 501 with the reason, not 404", async () => {
     for (const [method, path, feature] of [
       ["GET", "/api/v1/high-risk/status", "high-risk partitions"],
       ["GET", "/api/v1/agent/job-profiles", "agent job profiles"],
       ["POST", "/api/v1/agent/recovery/split", "agent recovery split"],
-      ["POST", "/api/v1/semantic/search", "semantic"],
     ] as const) {
       const res = await call(method, path);
       // 404 would read as "you typed the path wrong" and send an integrator
diff --git a/js/packages/core/src/index.ts b/js/packages/core/src/index.ts
index d6f61ea8..b07ca85b 100644
--- a/js/packages/core/src/index.ts
+++ b/js/packages/core/src/index.ts
@@ -88,6 +88,20 @@ export {
   SchemaMigrationError,
 } from "./schema/migrations.js";
 export { encodeQr, renderQrText, EC_LEVELS, type QrMatrix, type EcLevel } from "./util/qr.js";
+export {
+  buildSemanticRecords,
+  searchSemanticRecords,
+  semanticText,
+  semanticManifest,
+  semanticStatus,
+  tokenize,
+  SEMANTIC_KINDS,
+  SEMANTIC_MODEL_ID,
+  SEMANTIC_SCHEMA_VERSION,
+  type SemanticRecord,
+  type SemanticHit,
+  type SemanticStatus,
+} from "./vault/semanticIndex.js";
 export {
   findDerivationCollisions,
   APP32_KINDS,
diff --git a/js/packages/core/src/vault/semanticIndex.ts b/js/packages/core/src/vault/semanticIndex.ts
new file mode 100644
index 00000000..0fb8a64b
--- /dev/null
+++ b/js/packages/core/src/vault/semanticIndex.ts
@@ -0,0 +1,221 @@
+/**
+ * Local retrieval index over entry metadata — parity with
+ * src/seedpass/core/semantic_index.py.
+ *
+ * Despite the name this is not vector search and needs no model: it tokenizes
+ * an entry's descriptive fields and scores a query by Jaccard overlap
+ * (|intersection| / |union|) over those token sets. That is why it can be
+ * ported at all — there is nothing to embed.
+ *
+ * IT MUST NEVER INDEX SECRET MATERIAL. The records file is plaintext on disk,
+ * so anything indexed is readable by whatever can read the profile directory
+ * — a backup, a sync client, another user on a shared machine — without the
+ * master password. Python used to append a key_value entry's `value` here,
+ * which put the stored secret in that file in the clear along with a
+ * tokenized copy that leaked it just as well; that is fixed there and was
+ * never done here. The rule is simple and worth keeping simple: nobody
+ * searches for a secret they do not already know, so indexing one is pure
+ * downside.
+ */
+
+const WORD_RE = /[a-z0-9_]+/g;
+
+export const SEMANTIC_SCHEMA_VERSION = 1;
+/** Bumped when secret values stopped being indexed; a v1 index still holds them. */
+export const SEMANTIC_MODEL_ID = "seedpass-token-overlap-v2";
+
+/** Kinds worth indexing. Others carry nothing searchable that is not secret. */
+export const SEMANTIC_KINDS = new Set([
+  "document",
+  "note",
+  "key_value",
+  "password",
+  "stored_password",
+  "totp",
+  "nostr",
+  "ssh",
+  "pgp",
+]);
+
+export interface SemanticRecord {
+  entry_id: number;
+  kind: string;
+  label: string;
+  text: string;
+  /** Sorted, for a stable on-disk form. */
+  tokens: string[];
+}
+
+export interface SemanticHit {
+  entry_id: number;
+  kind: string;
+  label: string;
+  score: number;
+  excerpt: string;
+}
+
+export interface SemanticStatus {
+  enabled: boolean;
+  built: boolean;
+  records: number;
+  schema_version: number;
+  model_id: string;
+  updated_at: number;
+}
+
+export function tokenize(text: string): Set<string> {
+  const lowered = text.trim().toLowerCase();
+  if (!lowered) return new Set();
+  return new Set(lowered.match(WORD_RE) ?? []);
+}
+
+/**
+ * The searchable text for an entry.
+ *
+ * Every field here is a NAME or a note, never a value. Adding a field to this
+ * function puts it in a plaintext file, so the question to ask of any
+ * addition is not "would this be useful to search" but "would I be content
+ * for it to sit unencrypted next to the vault".
+ */
+export function semanticText(entry: Record<string, unknown>, kind: string): string {
+  const str = (v: unknown): string => String(v ?? "").trim();
+  const parts: string[] = [str(entry["label"]), str(entry["notes"])];
+
+  const tags = entry["tags"];
+  parts.push(
+    Array.isArray(tags) ? tags.map((t) => str(t)).filter(Boolean).join(" ") : str(tags),
+  );
+
+  if (kind === "document" || kind === "note") {
+    parts.push(str(entry["content"]), str(entry["file_type"]));
+  } else if (kind === "password" || kind === "stored_password") {
+    // Username and URL, never the derived password.
+    parts.push(str(entry["username"]), str(entry["url"]));
+  } else if (kind === "key_value") {
+    // The key NAME only. `value` is the secret; see the file header.
+    parts.push(str(entry["key"]));
+  } else if (kind === "totp") {
+    parts.push(str(entry["issuer"]));
+  } else if (kind === "nostr") {
+    // The public half only.
+    parts.push(str(entry["npub"]));
+  } else if (kind === "ssh" || kind === "pgp") {
+    parts.push(str(entry["fingerprint"]));
+  }
+
+  const links = entry["links"];
+  if (Array.isArray(links)) {
+    for (const link of links) {
+      if (typeof link !== "object" || link === null) continue;
+      const l = link as Record<string, unknown>;
+      parts.push(str(l["relation"]), str(l["note"]));
+    }
+  }
+
+  return parts.filter((p) => p).join("\n");
+}
+
+/** Build records from entries, skipping kinds and entries with nothing to index. */
+export function buildSemanticRecords(
+  entries: Array<Record<string, unknown>>,
+): SemanticRecord[] {
+  const records: SemanticRecord[] = [];
+  for (const entry of entries) {
+    // Entry 0 is a real entry — it is the FIRST one any profile creates — so
+    // it cannot double as "missing". Both implementations used to test
+    // `id <= 0`, which silently dropped it along with genuinely absent ids
+    // and left every user's first entry permanently unfindable, with nothing
+    // to indicate why. Fixed in both; see test/semanticIndex.test.ts.
+    const rawId = entry["id"];
+    if (rawId === undefined || rawId === null) continue;
+    const entryId = Number(rawId);
+    if (!Number.isInteger(entryId) || entryId < 0) continue;
+    const kind = String(entry["kind"] ?? entry["type"] ?? "").trim().toLowerCase();
+    if (!SEMANTIC_KINDS.has(kind)) continue;
+    const text = semanticText(entry, kind);
+    const tokens = tokenize(text);
+    if (!text.trim() || tokens.size === 0) continue;
+    records.push({
+      entry_id: entryId,
+      kind,
+      label: String(entry["label"] ?? ""),
+      text,
+      tokens: [...tokens].sort(),
+    });
+  }
+  return records;
+}
+
+/**
+ * Score a query against built records.
+ *
+ * Jaccard overlap, ties broken by entry id so the order is stable and matches
+ * Python's. Scores are rounded to six places for the same reason.
+ */
+export function searchSemanticRecords(
+  records: SemanticRecord[],
+  query: string,
+  options: { k?: number; kind?: string | null } = {},
+): SemanticHit[] {
+  if (records.length === 0) return [];
+  const queryTokens = tokenize(query);
+  if (queryTokens.size === 0) return [];
+  const wantedKind = options.kind ? options.kind.trim().toLowerCase() : null;
+
+  const scored: Array<{ score: number; record: SemanticRecord }> = [];
+  for (const record of records) {
+    const recordKind = String(record.kind ?? "").trim().toLowerCase();
+    if (wantedKind && recordKind !== wantedKind) continue;
+    const tokens = new Set(
+      record.tokens.map((t) => String(t).trim().toLowerCase()).filter((t) => t),
+    );
+    if (tokens.size === 0) continue;
+    let intersection = 0;
+    for (const token of queryTokens) if (tokens.has(token)) intersection++;
+    if (intersection <= 0) continue;
+    const union = new Set([...queryTokens, ...tokens]).size;
+    scored.push({ score: intersection / (union || 1), record });
+  }
+
+  scored.sort((a, b) => b.score - a.score || a.record.entry_id - b.record.entry_id);
+  const limit = Math.max(1, Math.trunc(options.k ?? 10));
+  return scored.slice(0, limit).map(({ score, record }) => ({
+    entry_id: record.entry_id,
+    kind: record.kind,
+    label: record.label,
+    // Six decimal places, as Python rounds it.
+    score: Math.round(score * 1e6) / 1e6,
+    excerpt: record.text.slice(0, 220),
+  }));
+}
+
+/** The manifest written beside the records. */
+export function semanticManifest(options: {
+  enabled: boolean;
+  built: boolean;
+  recordCount: number;
+  updatedAt: number;
+}): Record<string, unknown> {
+  return {
+    enabled: options.enabled,
+    built: options.built,
+    schema_version: SEMANTIC_SCHEMA_VERSION,
+    model_id: SEMANTIC_MODEL_ID,
+    updated_at: options.updatedAt,
+    record_count: options.recordCount,
+  };
+}
+
+export function semanticStatus(
+  manifest: Record<string, unknown>,
+  recordCount: number,
+): SemanticStatus {
+  return {
+    enabled: Boolean(manifest["enabled"] ?? false),
+    built: Boolean(manifest["built"] ?? false),
+    records: recordCount,
+    schema_version: Number(manifest["schema_version"] ?? SEMANTIC_SCHEMA_VERSION),
+    model_id: String(manifest["model_id"] ?? SEMANTIC_MODEL_ID),
+    updated_at: Number(manifest["updated_at"] ?? 0),
+  };
+}
diff --git a/js/packages/core/test/semanticIndex.test.ts b/js/packages/core/test/semanticIndex.test.ts
new file mode 100644
index 00000000..14731df1
--- /dev/null
+++ b/js/packages/core/test/semanticIndex.test.ts
@@ -0,0 +1,143 @@
+/**
+ * The retrieval index, checked against the Python implementation.
+ *
+ * Two properties matter, and they pull in different directions:
+ *
+ *   1. It must rank identically to Python, so a query answered on one
+ *      implementation is answered the same way on the other. Fixtures below
+ *      are that implementation's actual output.
+ *   2. It must never write a secret. The records file is plaintext on disk,
+ *      so anything indexed is readable by whatever can read the profile
+ *      directory without the master password. Python used to index a
+ *      key_value entry's `value`; the fixture proves it no longer does, and
+ *      the TS side is asserted against the same rule.
+ */
+
+import { describe, expect, it } from "vitest";
+import {
+  buildSemanticRecords,
+  searchSemanticRecords,
+  semanticStatus,
+  tokenize,
+  SEMANTIC_MODEL_ID,
+  type SemanticRecord,
+} from "@seedpass/core";
+import { semanticFixture } from "@seedpass/test-vectors";
+
+// Imported through the fixture package rather than read from disk, so the
+// suite also runs under the jsdom environment, where there is no filesystem.
+const fx = semanticFixture;
+
+describe("record building matches Python", () => {
+  it("produces the same records, in the same order, with the same tokens", () => {
+    expect(buildSemanticRecords(fx.entries)).toEqual(fx.records);
+  });
+
+  it("indexes entry 0, and skips only what genuinely has nothing to index", () => {
+    const ids = buildSemanticRecords(fx.entries).map((r) => r.entry_id);
+    // Entry 0 is the FIRST entry any profile creates. Both implementations
+    // used to test `id <= 0` and silently dropped it, so every user's first
+    // entry was permanently unfindable with nothing to indicate why.
+    expect(ids).toContain(0);
+    expect(ids).not.toContain(7); // 'seed' is not an indexed kind
+    expect(ids).not.toContain(8); // no text at all
+    expect(ids).toContain(9); // a link's relation and note are indexable text
+  });
+
+  it("still skips an entry with no id, which is what the old test meant", () => {
+    // The fixture carries one entry with no `id` key. Dropping it is correct;
+    // dropping id 0 alongside it was the bug.
+    const withoutId = fx.entries.filter((e) => e["id"] === undefined);
+    expect(withoutId).toHaveLength(1);
+    expect(buildSemanticRecords(withoutId)).toEqual([]);
+  });
+
+  it("finds the first entry a profile ever created", () => {
+    const hits = searchSemanticRecords(buildSemanticRecords(fx.entries), "first entry");
+    expect(hits.map((h) => h.entry_id)).toContain(0);
+  });
+});
+
+describe("no secret reaches the index", () => {
+  it("omits a key_value entry's value while keeping its key name", () => {
+    const records = buildSemanticRecords(fx.entries);
+    const serialized = JSON.stringify(records);
+    // The fixture entry carries this value; it must not survive indexing.
+    expect(serialized).not.toContain("SECRET-MUST-NOT-APPEAR");
+    // Tokenized fragments leak it just as well as the whole string.
+    for (const fragment of ["secret", "appear"]) {
+      expect(serialized.toLowerCase()).not.toContain(fragment);
+    }
+    // Still findable by the things that are not the secret.
+    const kv = records.find((r) => r.entry_id === 2)!;
+    expect(kv.text).toContain("DEPLOY_TOKEN");
+    expect(kv.text).toContain("ci pipeline");
+  });
+
+  it("indexes only the public half of a nostr entry", () => {
+    const nostr = buildSemanticRecords(fx.entries).find((r) => r.entry_id === 6)!;
+    expect(nostr.text).toContain("npub1example");
+    expect(nostr.text).not.toContain("nsec");
+  });
+});
+
+describe("search matches Python", () => {
+  for (const [query, expected] of Object.entries(fx.queries)) {
+    it(`ranks "${query}" identically`, () => {
+      const got = searchSemanticRecords(fx.records, query, { k: 10 });
+      expect(got).toEqual(expected);
+    });
+  }
+
+  it("filters by kind", () => {
+    const got = searchSemanticRecords(fx.records, "ops", { k: 10, kind: "ssh" });
+    expect(got.map((h) => h.entry_id)).toEqual(fx.kind_filtered.map((h) => h.entry_id));
+  });
+
+  it("returns nothing for an empty query rather than everything", () => {
+    // An empty query has no tokens, so nothing overlaps it. Returning the
+    // whole vault would be the dangerous reading of "matches everything".
+    expect(searchSemanticRecords(fx.records, "")).toEqual([]);
+    expect(searchSemanticRecords(fx.records, "   ")).toEqual([]);
+  });
+
+  it("honours k, keeping the highest scores", () => {
+    const all = searchSemanticRecords(fx.records, "bank account", { k: 10 });
+    const one = searchSemanticRecords(fx.records, "bank account", { k: 1 });
+    expect(one).toHaveLength(1);
+    expect(one[0]).toEqual(all[0]);
+  });
+
+  it("breaks score ties by entry id, so the order is stable", () => {
+    const records: SemanticRecord[] = [
+      { entry_id: 9, kind: "document", label: "b", text: "alpha", tokens: ["alpha"] },
+      { entry_id: 2, kind: "document", label: "a", text: "alpha", tokens: ["alpha"] },
+    ];
+    expect(searchSemanticRecords(records, "alpha").map((h) => h.entry_id)).toEqual([2, 9]);
+  });
+});
+
+describe("tokenizing", () => {
+  it("lowercases and splits on non-word characters", () => {
+    // Underscore is a word character, so `Login_2FA` is one token rather than
+    // two — verified against Python's [a-z0-9_]+ rather than assumed.
+    expect([...tokenize("Bank.Example/Login_2FA")].sort()).toEqual([
+      "bank",
+      "example",
+      "login_2fa",
+    ]);
+  });
+
+  it("is empty for whitespace", () => {
+    expect(tokenize("   ").size).toBe(0);
+  });
+});
+
+describe("status", () => {
+  it("reports the model id that marks a post-fix index", () => {
+    // A v1 index was built by code that wrote secrets, so the id has to
+    // distinguish them.
+    expect(semanticStatus({ built: true, enabled: true }, 3).model_id).toBe(SEMANTIC_MODEL_ID);
+    expect(SEMANTIC_MODEL_ID).toBe("seedpass-token-overlap-v2");
+  });
+});
diff --git a/js/packages/test-vectors/fixtures/manifest.json b/js/packages/test-vectors/fixtures/manifest.json
index d35ae088..2cdf6269 100644
--- a/js/packages/test-vectors/fixtures/manifest.json
+++ b/js/packages/test-vectors/fixtures/manifest.json
@@ -22,6 +22,7 @@
     "pgp_keys.json",
     "portable_backup.json",
     "qr.json",
+    "semantic.json",
     "ssh_keys.json",
     "sync_merge.json",
     "totp.json",
@@ -29,6 +30,6 @@
   ],
   "fixture_version": 1,
   "generator": "scripts/generate_ts_port_fixtures.py",
-  "python_commit": "46ff51f3d64a2e7a85deae90688b4312ab4f15fa",
+  "python_commit": "51f1358da4d9a743ca5d079c78b827fa84599c89",
   "redaction_policy": "All fixtures derive from public BIP-39 test mnemonics (abandon.../legal.../zoo...). Secret-shaped outputs are generated test material, never from a live profile."
 }
diff --git a/js/packages/test-vectors/fixtures/semantic.json b/js/packages/test-vectors/fixtures/semantic.json
new file mode 100644
index 00000000..e1bed35c
--- /dev/null
+++ b/js/packages/test-vectors/fixtures/semantic.json
@@ -0,0 +1,325 @@
+{
+  "description": "semantic index records and rankings",
+  "entries": [
+    {
+      "id": 0,
+      "kind": "password",
+      "label": "first-ever.example",
+      "notes": "the very first entry a profile creates",
+      "tags": [
+        "edge"
+      ],
+      "username": "zero"
+    },
+    {
+      "id": 1,
+      "kind": "password",
+      "label": "bank.example",
+      "notes": "main current account",
+      "tags": [
+        "money",
+        "daily"
+      ],
+      "url": "https://bank.example",
+      "username": "alice"
+    },
+    {
+      "id": 2,
+      "key": "DEPLOY_TOKEN",
+      "kind": "key_value",
+      "label": "deploy-token",
+      "notes": "ci pipeline",
+      "tags": [
+        "ops"
+      ],
+      "value": "SECRET-MUST-NOT-APPEAR"
+    },
+    {
+      "content": "how to recover the bank account if locked out",
+      "id": 3,
+      "kind": "document",
+      "label": "recovery notes",
+      "tags": [
+        "docs"
+      ]
+    },
+    {
+      "id": 4,
+      "issuer": "Fastmail",
+      "kind": "totp",
+      "label": "email-2fa",
+      "notes": "",
+      "tags": [
+        "email"
+      ]
+    },
+    {
+      "fingerprint": "SHA256:abcdef",
+      "id": 5,
+      "kind": "ssh",
+      "label": "prod-server",
+      "notes": "production access",
+      "tags": [
+        "ops"
+      ]
+    },
+    {
+      "id": 6,
+      "kind": "nostr",
+      "label": "social",
+      "notes": "",
+      "npub": "npub1example",
+      "tags": []
+    },
+    {
+      "id": 7,
+      "kind": "seed",
+      "label": "child seed",
+      "notes": "not an indexed kind"
+    },
+    {
+      "content": "",
+      "id": 8,
+      "kind": "document",
+      "label": "",
+      "notes": "",
+      "tags": []
+    },
+    {
+      "id": 9,
+      "kind": "password",
+      "label": "linked",
+      "links": [
+        {
+          "note": "see recovery notes",
+          "relation": "depends_on"
+        }
+      ],
+      "notes": "",
+      "tags": []
+    },
+    {
+      "content": "skipped",
+      "kind": "document",
+      "label": "no id at all"
+    }
+  ],
+  "fixture_version": 1,
+  "kind_filtered": [
+    {
+      "entry_id": 5,
+      "excerpt": "prod-server\nproduction access\nops\nSHA256:abcdef",
+      "kind": "ssh",
+      "label": "prod-server",
+      "score": 0.142857
+    }
+  ],
+  "queries": {
+    "DEPLOY_TOKEN": [
+      {
+        "entry_id": 2,
+        "excerpt": "deploy-token\nci pipeline\nops\nDEPLOY_TOKEN",
+        "kind": "key_value",
+        "label": "deploy-token",
+        "score": 0.166667
+      }
+    ],
+    "bank": [
+      {
+        "entry_id": 1,
+        "excerpt": "bank.example\nmain current account\nmoney daily\nalice\nhttps://bank.example",
+        "kind": "password",
+        "label": "bank.example",
+        "score": 0.111111
+      },
+      {
+        "entry_id": 3,
+        "excerpt": "recovery notes\ndocs\nhow to recover the bank account if locked out",
+        "kind": "document",
+        "label": "recovery notes",
+        "score": 0.083333
+      }
+    ],
+    "bank account": [
+      {
+        "entry_id": 1,
+        "excerpt": "bank.example\nmain current account\nmoney daily\nalice\nhttps://bank.example",
+        "kind": "password",
+        "label": "bank.example",
+        "score": 0.222222
+      },
+      {
+        "entry_id": 3,
+        "excerpt": "recovery notes\ndocs\nhow to recover the bank account if locked out",
+        "kind": "document",
+        "label": "recovery notes",
+        "score": 0.166667
+      }
+    ],
+    "email": [
+      {
+        "entry_id": 4,
+        "excerpt": "email-2fa\nemail\nFastmail",
+        "kind": "totp",
+        "label": "email-2fa",
+        "score": 0.333333
+      }
+    ],
+    "first entry": [
+      {
+        "entry_id": 0,
+        "excerpt": "first-ever.example\nthe very first entry a profile creates\nedge\nzero",
+        "kind": "password",
+        "label": "first-ever.example",
+        "score": 0.181818
+      }
+    ],
+    "nothing matches here": [],
+    "ops production": [
+      {
+        "entry_id": 5,
+        "excerpt": "prod-server\nproduction access\nops\nSHA256:abcdef",
+        "kind": "ssh",
+        "label": "prod-server",
+        "score": 0.285714
+      },
+      {
+        "entry_id": 2,
+        "excerpt": "deploy-token\nci pipeline\nops\nDEPLOY_TOKEN",
+        "kind": "key_value",
+        "label": "deploy-token",
+        "score": 0.142857
+      }
+    ],
+    "recover locked": [
+      {
+        "entry_id": 3,
+        "excerpt": "recovery notes\ndocs\nhow to recover the bank account if locked out",
+        "kind": "document",
+        "label": "recovery notes",
+        "score": 0.166667
+      }
+    ]
+  },
+  "records": [
+    {
+      "entry_id": 0,
+      "kind": "password",
+      "label": "first-ever.example",
+      "text": "first-ever.example\nthe very first entry a profile creates\nedge\nzero",
+      "tokens": [
+        "a",
+        "creates",
+        "edge",
+        "entry",
+        "ever",
+        "example",
+        "first",
+        "profile",
+        "the",
+        "very",
+        "zero"
+      ]
+    },
+    {
+      "entry_id": 1,
+      "kind": "password",
+      "label": "bank.example",
+      "text": "bank.example\nmain current account\nmoney daily\nalice\nhttps://bank.example",
+      "tokens": [
+        "account",
+        "alice",
+        "bank",
+        "current",
+        "daily",
+        "example",
+        "https",
+        "main",
+        "money"
+      ]
+    },
+    {
+      "entry_id": 2,
+      "kind": "key_value",
+      "label": "deploy-token",
+      "text": "deploy-token\nci pipeline\nops\nDEPLOY_TOKEN",
+      "tokens": [
+        "ci",
+        "deploy",
+        "deploy_token",
+        "ops",
+        "pipeline",
+        "token"
+      ]
+    },
+    {
+      "entry_id": 3,
+      "kind": "document",
+      "label": "recovery notes",
+      "text": "recovery notes\ndocs\nhow to recover the bank account if locked out",
+      "tokens": [
+        "account",
+        "bank",
+        "docs",
+        "how",
+        "if",
+        "locked",
+        "notes",
+        "out",
+        "recover",
+        "recovery",
+        "the",
+        "to"
+      ]
+    },
+    {
+      "entry_id": 4,
+      "kind": "totp",
+      "label": "email-2fa",
+      "text": "email-2fa\nemail\nFastmail",
+      "tokens": [
+        "2fa",
+        "email",
+        "fastmail"
+      ]
+    },
+    {
+      "entry_id": 5,
+      "kind": "ssh",
+      "label": "prod-server",
+      "text": "prod-server\nproduction access\nops\nSHA256:abcdef",
+      "tokens": [
+        "abcdef",
+        "access",
+        "ops",
+        "prod",
+        "production",
+        "server",
+        "sha256"
+      ]
+    },
+    {
+      "entry_id": 6,
+      "kind": "nostr",
+      "label": "social",
+      "text": "social\nnpub1example",
+      "tokens": [
+        "npub1example",
+        "social"
+      ]
+    },
+    {
+      "entry_id": 9,
+      "kind": "password",
+      "label": "linked",
+      "text": "linked\ndepends_on\nsee recovery notes",
+      "tokens": [
+        "depends_on",
+        "linked",
+        "notes",
+        "recovery",
+        "see"
+      ]
+    }
+  ]
+}
diff --git a/js/packages/test-vectors/src/index.ts b/js/packages/test-vectors/src/index.ts
index b0d41f1d..35944bee 100644
--- a/js/packages/test-vectors/src/index.ts
+++ b/js/packages/test-vectors/src/index.ts
@@ -9,6 +9,7 @@
 
 import manifestJson from "../fixtures/manifest.json";
 import qrJson from "../fixtures/qr.json";
+import semanticJson from "../fixtures/semantic.json";
 import bip39Json from "../fixtures/bip39_seeds.json";
 import bip85Json from "../fixtures/bip85_entropy.json";
 import passwordsV1Json from "../fixtures/passwords_v1.json";
@@ -315,3 +316,20 @@ export interface QrCase {
 }
 
 export const qrCases = (qrJson as { cases: QrCase[] }).cases;
+
+/** Semantic-index reference data, generated by the Python implementation. */
+export const semanticFixture = semanticJson as {
+  entries: Array<Record<string, unknown>>;
+  records: Array<{
+    entry_id: number;
+    kind: string;
+    label: string;
+    text: string;
+    tokens: string[];
+  }>;
+  queries: Record<
+    string,
+    Array<{ entry_id: number; kind: string; label: string; score: number; excerpt: string }>
+  >;
+  kind_filtered: Array<{ entry_id: number }>;
+};
diff --git a/scripts/generate_ts_port_fixtures.py b/scripts/generate_ts_port_fixtures.py
index 449bbfa3..d6c00ef8 100644
--- a/scripts/generate_ts_port_fixtures.py
+++ b/scripts/generate_ts_port_fixtures.py
@@ -1285,6 +1285,77 @@ def gen_qr() -> dict:
     }
 
 
+def gen_semantic() -> dict:
+    """Reference records and rankings for the retrieval index.
+
+    Two properties are being pinned. The rankings must match so a query
+    answered on one implementation is answered the same way on the other. And
+    the records must contain no secret: the file this produces is plaintext
+    beside an encrypted vault, so anything indexed is readable without the
+    master password. The key_value case below carries a value that must not
+    appear anywhere in the output.
+    """
+    import tempfile
+
+    from seedpass.core.semantic_index import SemanticIndex
+
+    entries = [
+        {"id": 0, "kind": "password", "label": "first-ever.example",
+         "username": "zero", "notes": "the very first entry a profile creates",
+         "tags": ["edge"]},
+        {"id": 1, "kind": "password", "label": "bank.example", "username": "alice",
+         "url": "https://bank.example", "notes": "main current account",
+         "tags": ["money", "daily"]},
+        {"id": 2, "kind": "key_value", "label": "deploy-token", "key": "DEPLOY_TOKEN",
+         "value": "SECRET-MUST-NOT-APPEAR", "notes": "ci pipeline", "tags": ["ops"]},
+        {"id": 3, "kind": "document", "label": "recovery notes",
+         "content": "how to recover the bank account if locked out", "tags": ["docs"]},
+        {"id": 4, "kind": "totp", "label": "email-2fa", "issuer": "Fastmail",
+         "notes": "", "tags": ["email"]},
+        {"id": 5, "kind": "ssh", "label": "prod-server", "fingerprint": "SHA256:abcdef",
+         "notes": "production access", "tags": ["ops"]},
+        {"id": 6, "kind": "nostr", "label": "social", "npub": "npub1example",
+         "notes": "", "tags": []},
+        {"id": 7, "kind": "seed", "label": "child seed", "notes": "not an indexed kind"},
+        {"id": 8, "kind": "document", "label": "", "content": "", "notes": "", "tags": []},
+        {"id": 9, "kind": "password", "label": "linked", "notes": "", "tags": [],
+         "links": [{"relation": "depends_on", "note": "see recovery notes"}]},
+        {"kind": "document", "label": "no id at all", "content": "skipped"},
+    ]
+    queries = [
+        "bank",
+        "bank account",
+        "ops production",
+        "email",
+        "recover locked",
+        "nothing matches here",
+        "DEPLOY_TOKEN",
+        "first entry",
+    ]
+
+    with tempfile.TemporaryDirectory() as tmpdir:
+        index = SemanticIndex(Path(tmpdir))
+        index.build(entries)
+        records = json.loads(
+            (Path(tmpdir) / "semantic_index" / "records.json").read_text(
+                encoding="utf-8"
+            )
+        )
+        results = {q: index.search(q, k=10) for q in queries}
+        kind_filtered = index.search("ops", k=10, kind="ssh")
+
+    blob = json.dumps(records)
+    assert "SECRET-MUST-NOT-APPEAR" not in blob, "the index leaked a stored secret"
+
+    return {
+        "description": "semantic index records and rankings",
+        "entries": entries,
+        "records": records,
+        "queries": results,
+        "kind_filtered": kind_filtered,
+    }
+
+
 def main() -> None:
     FIXTURES_DIR.mkdir(parents=True, exist_ok=True)
 
@@ -1309,6 +1380,7 @@ def main() -> None:
         "password_kdf.json": gen_password_kdf(),
         "legacy_payloads.json": gen_legacy_payloads(),
         "qr.json": gen_qr(),
+        "semantic.json": gen_semantic(),
     }
     entries_fixture, vault_fixture = gen_entries_and_vault()
     files["entries_index.json"] = entries_fixture
diff --git a/src/seedpass/core/semantic_index.py b/src/seedpass/core/semantic_index.py
index cff601c5..4448674c 100644
--- a/src/seedpass/core/semantic_index.py
+++ b/src/seedpass/core/semantic_index.py
@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import json
+import os
 import re
 import time
 from dataclasses import dataclass
@@ -26,7 +27,10 @@ class SemanticIndex:
     MANIFEST_FILENAME = "manifest.json"
     RECORDS_FILENAME = "records.json"
     SCHEMA_VERSION = 1
-    MODEL_ID = "seedpass-token-overlap-v1"
+    #: Bumped from v1 when secret values stopped being indexed. An index
+    #: built by the old code still contains them, so `status()` reports the
+    #: stored id and a mismatch means "rebuild this, it holds secrets".
+    MODEL_ID = "seedpass-token-overlap-v2"
 
     ALLOWED_KINDS = {
         "document",
@@ -85,6 +89,10 @@ class SemanticIndex:
             json.dumps(serializable, ensure_ascii=True, sort_keys=True, indent=2),
             encoding="utf-8",
         )
+        # Labels, notes, usernames and URLs are still vault contents even with
+        # secret values excluded. Every other file in a profile is 0600; this
+        # one was created at the process umask (0664 on a default install).
+        os.chmod(self.records_path, 0o600)
         manifest = self._load_manifest()
         manifest.update(
             {
@@ -157,8 +165,20 @@ class SemanticIndex:
     ) -> list[SemanticRecord]:
         records: list[SemanticRecord] = []
         for entry in entries:
-            entry_id = int(entry.get("id", 0) or 0)
-            if entry_id <= 0:
+            # `id` 0 is a real entry -- it is the FIRST one any profile
+            # creates -- so it cannot double as "missing". The old test was
+            # `int(entry.get("id", 0) or 0) <= 0`, which silently dropped it
+            # along with genuinely absent ids, leaving every user's first
+            # entry permanently unfindable by search with nothing to indicate
+            # why.
+            raw_id = entry.get("id")
+            if raw_id is None:
+                continue
+            try:
+                entry_id = int(raw_id)
+            except (TypeError, ValueError):
+                continue
+            if entry_id < 0:
                 continue
             kind = str(entry.get("kind") or entry.get("type") or "").strip().lower()
             if kind not in self.ALLOWED_KINDS:
@@ -197,8 +217,15 @@ class SemanticIndex:
             parts.append(str(entry.get("username", "")).strip())
             parts.append(str(entry.get("url", "")).strip())
         elif kind == "key_value":
+            # The key NAME only. `value` is the stored secret itself, and it
+            # used to be indexed here -- which wrote it, in the clear, to
+            # semantic_index/records.json alongside a tokenized copy. Nobody
+            # searches for a secret they do not already know, so indexing it
+            # was pure downside: every other file in the profile is encrypted
+            # and 0600, and this one handed the secret to anything that could
+            # read the directory (a backup, a sync client, another user on a
+            # shared box) without the master password.
             parts.append(str(entry.get("key", "")).strip())
-            parts.append(str(entry.get("value", "")).strip())
         elif kind == "totp":
             parts.append(str(entry.get("issuer", "")).strip())
         elif kind == "nostr":
@@ -238,6 +265,7 @@ class SemanticIndex:
             json.dumps(manifest, ensure_ascii=True, sort_keys=True, indent=2),
             encoding="utf-8",
         )
+        os.chmod(self.manifest_path, 0o600)
 
     def _load_records(self) -> list[dict[str, Any]]:
         if not self.records_path.exists():
diff --git a/src/tests/test_semantic_index.py b/src/tests/test_semantic_index.py
index 6aedc675..25aef367 100644
--- a/src/tests/test_semantic_index.py
+++ b/src/tests/test_semantic_index.py
@@ -124,3 +124,101 @@ def test_semantic_index_service_build_and_status(tmp_path: Path) -> None:
 
     hybrid = service.search("agent", k=3, mode="hybrid")
     assert len(hybrid) >= 1
+
+
+def test_stored_secrets_are_never_written_to_the_index(tmp_path):
+    """The index is a plaintext file; secrets must not be in it.
+
+    `_extract_text` used to append a key_value entry's `value` -- the stored
+    secret itself -- which `build()` then wrote to
+    semantic_index/records.json in the clear, alongside a tokenized copy that
+    leaked it just as thoroughly. Nobody searches for a secret they do not
+    already know, so indexing it was pure downside: it handed the value to
+    anything that could read the profile directory (a backup, a sync client,
+    another user on a shared machine) without the master password.
+    """
+    from seedpass.core.semantic_index import SemanticIndex
+
+    index = SemanticIndex(tmp_path)
+    index.build(
+        [
+            {
+                "id": 1,
+                "kind": "key_value",
+                "label": "api-token",
+                "key": "DEPLOY_TOKEN",
+                "value": "SUPER-SECRET-VALUE-12345",
+                "notes": "production credentials",
+                "tags": ["ops"],
+            }
+        ]
+    )
+
+    raw = (tmp_path / "semantic_index" / "records.json").read_text(encoding="utf-8")
+    assert "SUPER-SECRET-VALUE-12345" not in raw
+    # Tokenized fragments leak it just as well as the whole string.
+    for fragment in ("super", "12345"):
+        assert fragment not in raw.lower()
+
+    # The entry is still findable by everything that is not the secret.
+    assert index.search("token")
+    assert index.search("production")
+    assert "DEPLOY_TOKEN" in raw
+
+
+def test_index_files_are_not_readable_by_other_users(tmp_path):
+    """Every other file in a profile is 0600; these were created at the umask."""
+    from seedpass.core.semantic_index import SemanticIndex
+
+    index = SemanticIndex(tmp_path)
+    index.set_enabled(True)
+    index.build([{"id": 1, "kind": "document", "label": "notes", "content": "hello"}])
+    for name in ("records.json", "manifest.json"):
+        mode = (tmp_path / "semantic_index" / name).stat().st_mode & 0o777
+        assert mode == 0o600, f"{name} is {oct(mode)}"
+
+
+def test_model_id_marks_indexes_built_before_the_fix(tmp_path):
+    """A stale index still contains secrets, so it has to be distinguishable."""
+    from seedpass.core.semantic_index import SemanticIndex
+
+    index = SemanticIndex(tmp_path)
+    index.build([{"id": 1, "kind": "document", "label": "notes", "content": "hello"}])
+    assert index.status()["model_id"] == "seedpass-token-overlap-v2"
+
+
+def test_the_first_entry_a_profile_creates_is_searchable(tmp_path):
+    """Entry 0 is a real entry, not a missing one.
+
+    The check was `int(entry.get("id", 0) or 0) <= 0`, which conflated "no
+    id" with "id 0" — and entry ids start at 0, so the first entry any user
+    ever created was silently absent from every search result, with nothing
+    to indicate why.
+    """
+    from seedpass.core.semantic_index import SemanticIndex
+
+    index = SemanticIndex(tmp_path)
+    index.build(
+        [
+            {"id": 0, "kind": "password", "label": "bank.example", "notes": "first"},
+            {"id": 1, "kind": "password", "label": "forum.example", "notes": "second"},
+        ]
+    )
+    assert [hit["entry_id"] for hit in index.search("bank")] == [0]
+    assert [hit["entry_id"] for hit in index.search("forum")] == [1]
+
+
+def test_entries_without_a_usable_id_are_still_skipped(tmp_path):
+    """Which is what the old check was reaching for."""
+    from seedpass.core.semantic_index import SemanticIndex
+
+    index = SemanticIndex(tmp_path)
+    index.build(
+        [
+            {"kind": "document", "label": "no id", "content": "text"},
+            {"id": None, "kind": "document", "label": "null id", "content": "text"},
+            {"id": "abc", "kind": "document", "label": "bad id", "content": "text"},
+            {"id": -1, "kind": "document", "label": "negative", "content": "text"},
+        ]
+    )
+    assert index.search("text") == []