From 524a7cce3213acdd5db5a39ba596db26ca78ff59 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Tue, 27 Oct 2026 04:19:22 -0700
Subject: [PATCH] fix(native-agents): move summarizer, welcomer and clerk off
 the expired nemotron route

The free OpenRouter nemotron key expired on 2026-10-25. LiteLLM still
declares the name, so nothing failed at boot, but every model call returns
HTTP 404: pod-summarizer failed 6 of 8 runs in the 48h before 2026-10-27,
dying on its first model call. pod-welcomer and task-clerk carry the same
string and fail on their next trigger.

Point all three at deepseek-v4-flash, the model Scout and Recorder already
run on (Scout: 44 of 47 runs succeeded in the same window). The boot seed
$sets the demo-pod installs; other installs need
refresh-native-agent-configs after deploy.

The new test pins that every first-party app names a model LiteLLM declares
and does not route through the OpenRouter free tier, since a declared name
was not enough to catch this.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MDs9DS3E4dzoHKdxtgQqyz
---
 .../config/nativeAgents.modelRoute.test.js    | 35 +++++++++++++++++++
 .../config/native-agents/pod-summarizer.ts    |  2 +-
 backend/config/native-agents/pod-welcomer.ts  |  2 +-
 backend/config/native-agents/task-clerk.ts    |  2 +-
 4 files changed, 38 insertions(+), 3 deletions(-)
 create mode 100644 backend/__tests__/unit/config/nativeAgents.modelRoute.test.js

diff --git a/backend/__tests__/unit/config/nativeAgents.modelRoute.test.js b/backend/__tests__/unit/config/nativeAgents.modelRoute.test.js
new file mode 100644
index 0000000000..21fa97c755
--- /dev/null
+++ b/backend/__tests__/unit/config/nativeAgents.modelRoute.test.js
@@ -0,0 +1,35 @@
+/**
+ * First-party native apps must name a model the deployment can actually serve.
+ *
+ * pod-summarizer, pod-welcomer and task-clerk kept the free OpenRouter
+ * nemotron route after its key expired on 2026-10-25. LiteLLM still declares
+ * that name, so nothing failed at boot; every run died on its first model call
+ * with HTTP 404 (6 of 8 pod-summarizer runs in the 48h before 2026-10-27).
+ * Declared is not the same as servable, so this pins both.
+ */
+const fs = require('fs');
+const path = require('path');
+const { FIRST_PARTY_APPS } = require('../../../config/native-agents/apps');
+
+const litellmConfig = fs.readFileSync(path.join(__dirname,
+  '../../../../k8s/helm/commonly/templates/configmaps/litellm-config.yaml'), 'utf8');
+const declaredModels = new Set(
+  [...litellmConfig.matchAll(/model_name:\s*(\S+)/g)].map((m) => m[1]),
+);
+
+describe('first-party native app model routes', () => {
+  test.each(FIRST_PARTY_APPS.map((app) => [app.agentName, app.model]))(
+    '%s names a model LiteLLM declares',
+    (_name, model) => {
+      expect(declaredModels.has(model)).toBe(true);
+    },
+  );
+
+  test.each(FIRST_PARTY_APPS.map((app) => [app.agentName, app.model]))(
+    '%s does not route through the expired OpenRouter free tier',
+    (_name, model) => {
+      expect(model).not.toMatch(/^openrouter\//);
+      expect(model).not.toMatch(/:free$/);
+    },
+  );
+});
diff --git a/backend/config/native-agents/pod-summarizer.ts b/backend/config/native-agents/pod-summarizer.ts
index 2ea2e75d56..eb57d3a020 100644
--- a/backend/config/native-agents/pod-summarizer.ts
+++ b/backend/config/native-agents/pod-summarizer.ts
@@ -61,7 +61,7 @@ export const podSummarizerApp = {
     + '- DO NOT call commonly_create_task. You are not here to create work.',
   // See pod-welcomer.ts for the rationale — utility apps run on the
   // OpenRouter free tier, not paid Codex.
-  model: 'openrouter/nvidia/nemotron-3-super-120b-a12b:free',
+  model: 'deepseek-v4-flash',
   triggers: ['heartbeat'],
   heartbeatIntervalMinutes: 360,
   tools: [
diff --git a/backend/config/native-agents/pod-welcomer.ts b/backend/config/native-agents/pod-welcomer.ts
index 9f657cd9d4..6049451265 100644
--- a/backend/config/native-agents/pod-welcomer.ts
+++ b/backend/config/native-agents/pod-welcomer.ts
@@ -56,7 +56,7 @@ export const podWelcomerApp = {
   // burning it for these heavy-traffic event handlers exhausts dev agents'
   // accounts. Same default community agents use; if quality drops, reach for
   // gemini-2.5-flash (paid but cheap) before paid Codex.
-  model: 'openrouter/nvidia/nemotron-3-super-120b-a12b:free',
+  model: 'deepseek-v4-flash',
   triggers: ['pod.join'],
   tools: ['commonly_read_context', 'commonly_post_message'],
   iconUrl: '',
diff --git a/backend/config/native-agents/task-clerk.ts b/backend/config/native-agents/task-clerk.ts
index 69782f1332..d8c916c3fe 100644
--- a/backend/config/native-agents/task-clerk.ts
+++ b/backend/config/native-agents/task-clerk.ts
@@ -56,7 +56,7 @@ export const taskClerkApp = {
     + '- Keep all replies under 2 lines. You are a clerk, not a conversationalist.',
   // See pod-welcomer.ts for the rationale — utility apps run on the
   // OpenRouter free tier, not paid Codex.
-  model: 'openrouter/nvidia/nemotron-3-super-120b-a12b:free',
+  model: 'deepseek-v4-flash',
   triggers: ['mention'],
   tools: [
     'commonly_read_context',