diff --git a/.env.example b/.env.example
index 5ba3fed9..962e1bcf 100644
--- a/.env.example
+++ b/.env.example
@@ -83,10 +83,10 @@
 # AI_HISTORY_MAX_TURNS=10              # Older turns dropped (1–100)
 # AI_HISTORY_MAX_TURN_CHARS=4096       # Per-turn char cap (100–100000)
 # AI_HISTORY_MAX_TOTAL_CHARS=32768     # Total cap across all turns (100–1000000)
-# Feature flag: when true, the inventory query path matches bins server-side
-# and the LLM only formats the answer + relevance strings. Default false until
-# validated.
-# AI_DETERMINISTIC_MATCH=false
+# Inventory query: when true (default), uses LLM planner + SQL executor.
+# When false, falls back to legacy LLM-as-matcher. Disable for providers
+# without structured-output support (some Ollama models).
+# AI_DETERMINISTIC_MATCH=true
 
 # ── Backups ───────────────────────────────────
 # BACKUP_ENABLED=false
diff --git a/server/src/__tests__/aiStreamDeterministicFlag.test.ts b/server/src/__tests__/aiStreamDeterministicFlag.test.ts
index 4818c846..cd35911f 100644
--- a/server/src/__tests__/aiStreamDeterministicFlag.test.ts
+++ b/server/src/__tests__/aiStreamDeterministicFlag.test.ts
@@ -9,7 +9,7 @@ beforeEach(() => { app = createApp(); });
 afterEach(() => { delete process.env.AI_DETERMINISTIC_MATCH; });
 
 describe('AI_DETERMINISTIC_MATCH flag', () => {
-  it('legacy path is used when the flag is unset (no AI mock — just verifies the route still accepts the request)', async () => {
+  it('planner path is used by default (no AI mock — just verifies the route still accepts the request)', async () => {
     const { token } = await createTestUser(app);
     const loc = await createTestLocation(app, token);
     await createTestBin(app, token, loc.id, { name: 'Battery Bin' });
diff --git a/server/src/lib/__tests__/configDeterministicMatch.test.ts b/server/src/lib/__tests__/configDeterministicMatch.test.ts
index 503ed2b5..90543447 100644
--- a/server/src/lib/__tests__/configDeterministicMatch.test.ts
+++ b/server/src/lib/__tests__/configDeterministicMatch.test.ts
@@ -14,16 +14,16 @@ describe('config.aiDeterministicMatch', () => {
     await initialize();
   });
 
-  it('defaults to false when env var unset', async () => {
+  it('defaults to true when env var unset', async () => {
     delete process.env.AI_DETERMINISTIC_MATCH;
     const { config } = await import('../config.js');
-    expect(config.aiDeterministicMatch).toBe(false);
+    expect(config.aiDeterministicMatch).toBe(true);
   });
 
-  it('is true when AI_DETERMINISTIC_MATCH=true', async () => {
-    process.env.AI_DETERMINISTIC_MATCH = 'true';
+  it('is false when AI_DETERMINISTIC_MATCH=false', async () => {
+    process.env.AI_DETERMINISTIC_MATCH = 'false';
     const { config } = await import('../config.js');
-    expect(config.aiDeterministicMatch).toBe(true);
+    expect(config.aiDeterministicMatch).toBe(false);
     delete process.env.AI_DETERMINISTIC_MATCH;
   });
 });
diff --git a/server/src/lib/config.ts b/server/src/lib/config.ts
index cd0fe6bc..1041a406 100644
--- a/server/src/lib/config.ts
+++ b/server/src/lib/config.ts
@@ -250,10 +250,10 @@ export const config = Object.freeze({
   aiHistoryMaxTurns: clamp(parseInt(process.env.AI_HISTORY_MAX_TURNS || '10', 10), 1, 100, 10),
   aiHistoryMaxTurnChars: clamp(parseInt(process.env.AI_HISTORY_MAX_TURN_CHARS || '4096', 10), 100, 100000, 4096),
   aiHistoryMaxTotalChars: clamp(parseInt(process.env.AI_HISTORY_MAX_TOTAL_CHARS || '32768', 10), 100, 1_000_000, 32768),
-  // Feature flag: when true, the inventory query path uses deterministic
-  // server-side bin matching and reduces the LLM to a formatter. Default off
-  // — flip on per-environment to validate before retiring the legacy path.
-  aiDeterministicMatch: parseBool(process.env.AI_DETERMINISTIC_MATCH, false),
+  // Inventory query: when true (default), uses LLM planner + SQL executor.
+  // When false, falls back to legacy LLM-as-matcher. Disable for providers
+  // without structured-output support (some Ollama models).
+  aiDeterministicMatch: parseBool(process.env.AI_DETERMINISTIC_MATCH, true),
 
   // Backup
   backupEnabled: parseBool(process.env.BACKUP_ENABLED, false),