fix: recall via /v3/search so self-hosted backends inject matches

plugin/hooks/lib/api.js

@@ -51,6 +51,81 @@ function getProfile(baseUrl, apiKey, containerTag, query, options = {}) {
   return post(baseUrl, apiKey, '/v4/profile', { containerTag, q: query }, options.timeoutMs);
 }
 
+// Self-hosted backends sometimes return `score` instead of `similarity`, and
+// document-mode hits may nest chunk text under `chunks[]`. Normalize so the
+// recall hook's resultText/similarity filter sees a flat, consistent shape.
+function normalizeSearchHit(hit) {
+  if (!hit || typeof hit !== 'object') return null;
+  const similarity = Number.isFinite(hit.similarity)
+    ? hit.similarity
+    : Number.isFinite(hit.score)
+      ? hit.score
+      : undefined;
+  const filepath =
+    (typeof hit.filepath === 'string' && hit.filepath) ||
+    (typeof hit.metadata?.filepath === 'string' && hit.metadata.filepath) ||
+    undefined;
+  return {
+    ...hit,
+    ...(similarity !== undefined ? { similarity } : {}),
+    ...(filepath ? { filepath } : {}),
+  };
+}
+
+function flattenSearchResults(response) {
+  const raw = Array.isArray(response?.results) ? response.results : [];
+  const out = [];
+  for (const item of raw) {
+    if (Array.isArray(item?.chunks) && item.chunks.length > 0) {
+      for (const chunk of item.chunks) {
+        const text =
+          [chunk?.chunk, chunk?.content, chunk?.text, chunk?.memory].find(
+            (v) => typeof v === 'string' && v.trim(),
+          ) || null;
+        const normalized = normalizeSearchHit({
+          ...chunk,
+          ...(text && !chunk.chunk && !chunk.memory ? { chunk: text } : {}),
+          title: chunk.title || item.title,
+          filepath: chunk.filepath || item.filepath,
+          similarity: chunk.similarity ?? chunk.score ?? item.similarity ?? item.score,
+        });
+        if (normalized) out.push(normalized);
+      }
+      continue;
+    }
+    const normalized = normalizeSearchHit(item);
+    if (normalized) out.push(normalized);
+  }
+  return out;
+}
+
+// Per-prompt recall must not depend on /v4/profile's embedded searchResults —
+// on self-hosted backends that field stays empty even when /v3/search finds
+// real hits (issue #106). Dedicated search keeps cloud and local recall working.
+async function searchMemory(baseUrl, apiKey, containerTag, query, options = {}) {
+  const body = {
+    q: query,
+    // Singular is current; plural is what the #106 self-hosted repro used on
+    // /v3/search. Send both so neither cloud nor local silently scopes wrong.
+    containerTag,
+    containerTags: [containerTag],
+    limit: options.limit ?? 10,
+    searchMode: options.searchMode ?? 'hybrid',
+  };
+  const response = await post(
+    baseUrl,
+    apiKey,
+    '/v3/search',
+    body,
+    options.timeoutMs,
+  );
+  return {
+    results: flattenSearchResults(response),
+    total: response?.total,
+    timing: response?.timing,
+  };
+}
+
 function addMemory(baseUrl, apiKey, content, containerTag, metadata, options = {}) {
   const body = {
     content,
@@ -62,4 +137,9 @@ function addMemory(baseUrl, apiKey, content, containerTag, metadata, options = {
   return post(baseUrl, apiKey, '/v3/documents', body, options.timeoutMs);
 }
 
-module.exports = { AGENT_ENTITY_CONTEXT, getProfile, addMemory };
+module.exports = {
+  AGENT_ENTITY_CONTEXT,
+  getProfile,
+  searchMemory,
+  addMemory,
+};

plugin/hooks/recall-directive.js

@@ -2,7 +2,7 @@ const crypto = require('node:crypto');
 const fs = require('node:fs');
 const path = require('node:path');
 
-const { getProfile } = require('./lib/api');
+const { searchMemory } = require('./lib/api');
 const { BRAND, gray, red } = require('./lib/colors');
 const { getContainerTag } = require('./lib/container-tag');
 const { getUserFriendlyError } = require('./lib/error-helpers');
@@ -123,15 +123,17 @@ async function main() {
     }
 
     const containerTag = getContainerTag(cwd);
-    const response = await getProfile(
+    // Use /v3/search directly — /v4/profile's embedded searchResults is empty
+    // on self-hosted backends even when search finds hits (issue #106).
+    const response = await searchMemory(
       getBaseUrl(cwd, projectConfig),
       apiKey,
       containerTag,
       prompt.slice(0, MAX_QUERY_LENGTH),
-      { timeoutMs: SEARCH_TIMEOUT_MS },
+      { timeoutMs: SEARCH_TIMEOUT_MS, limit: MAX_RESULTS },
     );
 
-    const results = (response?.searchResults?.results || [])
+    const results = (response?.results || [])
       .filter((r) => resultText(r))
       .filter((r) => !Number.isFinite(r.similarity) || r.similarity >= MIN_SIMILARITY)
       .slice(0, MAX_RESULTS);

test/unit.mjs

@@ -177,15 +177,13 @@ describe('recall-directive hook', () => {
       res.setHeader('Content-Type', 'application/json');
       res.end(
         JSON.stringify({
-          searchResults: {
-            results: [
-              { memory: 'Chose Drizzle over Prisma', similarity: 0.82 },
-              { chunk: 'export const db = drizzle(client)', filepath: 'src/db.ts', similarity: 0.74 },
-              { memory: 'Errors must be loud and obvious', similarity: 0.71 },
-              { title: 'Migration plan', content: 'Use expand-contract migrations', similarity: 0.7 },
-              { memory: 'irrelevant low-similarity hit', similarity: 0.2 },
-            ],
-          },
+          results: [
+            { memory: 'Chose Drizzle over Prisma', similarity: 0.82 },
+            { chunk: 'export const db = drizzle(client)', filepath: 'src/db.ts', similarity: 0.74 },
+            { memory: 'Errors must be loud and obvious', similarity: 0.71 },
+            { title: 'Migration plan', content: 'Use expand-contract migrations', similarity: 0.7 },
+            { memory: 'irrelevant low-similarity hit', similarity: 0.2 },
+          ],
         }),
       );
     });
@@ -207,11 +205,11 @@ describe('recall-directive hook', () => {
     assert.doesNotMatch(context, /irrelevant low-similarity hit/);
     assert.match(context, /repo_example_project__/);
     assert.match(plain(output.systemMessage), /^◪ supermemory · recalled \d+ memories \(\d+ tok\)$/);
-    assert.equal(stub.requests[0].url, '/v4/profile');
-    assert.equal(
-      JSON.parse(stub.requests[0].body).q,
-      'continue the database work from before',
-    );
+    assert.equal(stub.requests[0].url, '/v3/search');
+    const requestBody = JSON.parse(stub.requests[0].body);
+    assert.equal(requestBody.q, 'continue the database work from before');
+    assert.match(requestBody.containerTag, /^repo_example_project__/);
+    assert.deepEqual(requestBody.containerTags, [requestBody.containerTag]);
 
     const state = readState('s1', {
       dataDir: join(home, '.supermemory-claude', 'statusline'),
@@ -253,7 +251,7 @@ describe('recall-directive hook', () => {
     ];
     const stub = await startStubServer(t, (record, res) => {
       res.setHeader('Content-Type', 'application/json');
-      res.end(JSON.stringify({ searchResults: { results: hits } }));
+      res.end(JSON.stringify({ results: hits }));
     });
     const env = { HOME: home, USERPROFILE: home, SUPERMEMORY_API_URL: stub.url };
     const input = { session_id: 's-dedup', cwd: repo, prompt: 'continue the database work' };
@@ -294,6 +292,57 @@ describe('recall-directive hook', () => {
     );
     assert.equal(JSON.parse(stdout).hookSpecificOutput.additionalContext, 'CUSTOM DIRECTIVE');
   });
+
+  test('recalls via /v3/search including score and nested chunk shapes (issue #106)', async (t) => {
+    const { repo, home } = makeRepo(t);
+    mkdirSync(join(home, '.supermemory-claude'), { recursive: true });
+    writeFileSync(
+      join(home, '.supermemory-claude', 'credentials.json'),
+      JSON.stringify({ apiKey: 'sm_test_key_0123456789abcdef' }),
+    );
+    const stub = await startStubServer(t, (record, res) => {
+      assert.equal(record.url, '/v3/search');
+      res.setHeader('Content-Type', 'application/json');
+      // Self-hosted shape from #106: score instead of similarity, and a
+      // document hit carrying nested chunks (total counts chunks, results
+      // length can be 1).
+      res.end(
+        JSON.stringify({
+          results: [
+            {
+              title: 'Region runbook',
+              score: 0.767,
+              chunks: [
+                { content: 'Failover steps for multi-region deploy', score: 0.767 },
+                { chunk: 'Health checks must probe both regions', score: 0.71 },
+              ],
+            },
+            { memory: 'too weak to inject', score: 0.2 },
+          ],
+          total: 3,
+        }),
+      );
+    });
+
+    const { code, stdout } = await runHook(
+      'recall-directive.js',
+      {
+        session_id: 's-v3',
+        cwd: repo,
+        prompt: 'how can I see all the things supermemory have saved already',
+      },
+      { HOME: home, USERPROFILE: home, SUPERMEMORY_API_URL: stub.url },
+    );
+    assert.equal(code, 0);
+    const output = JSON.parse(stdout);
+    const context = output.hookSpecificOutput.additionalContext;
+    assert.match(context, /Failover steps for multi-region deploy/);
+    assert.match(context, /Health checks must probe both regions/);
+    assert.doesNotMatch(context, /too weak to inject/);
+    assert.match(plain(output.systemMessage), /^◪ supermemory · recalled 2 memories \(\d+ tok\)$/);
+    assert.equal(stub.requests.length, 1);
+    assert.equal(stub.requests[0].url, '/v3/search');
+  });
 });
 
 describe('stdin handling', () => {