diff --git a/.github/workflows/nightly-e2e.yaml b/.github/workflows/nightly-e2e.yaml
index 06aaeab8db..280daa0b52 100644
--- a/.github/workflows/nightly-e2e.yaml
+++ b/.github/workflows/nightly-e2e.yaml
@@ -956,6 +956,7 @@ jobs:
       - name: Run token rotation E2E test
         env:
           NVIDIA_INFERENCE_API_KEY: ${{ secrets.NVIDIA_INFERENCE_API_KEY }}
+          NEMOCLAW_E2E_USE_NVIDIA_SECRET_AS_COMPATIBLE: "1"
           NEMOCLAW_NON_INTERACTIVE: "1"
           NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
           NEMOCLAW_POLICY_TIER: "open"
diff --git a/test/e2e-script-workflow.test.ts b/test/e2e-script-workflow.test.ts
index f75dbf9629..e5a2aab373 100644
--- a/test/e2e-script-workflow.test.ts
+++ b/test/e2e-script-workflow.test.ts
@@ -552,6 +552,18 @@ describe("E2E reusable workflow contract", () => {
     }
   });
 
+  it("routes legacy token rotation through the CI compatible inference endpoint", () => {
+    const runStep = nightlyWorkflow.jobs["token-rotation-e2e"].steps?.find(
+      (step) => step.name === "Run token rotation E2E test",
+    );
+    const script = readFileSync(new URL("./e2e/test-token-rotation.sh", import.meta.url), "utf8");
+
+    expect(runStep?.env?.NVIDIA_INFERENCE_API_KEY).toBe("${{ secrets.NVIDIA_INFERENCE_API_KEY }}");
+    expect(runStep?.env?.NEMOCLAW_E2E_USE_NVIDIA_SECRET_AS_COMPATIBLE).toBe("1");
+    expect(script).toContain("lib/ci-compatible-inference.sh");
+    expect(script).toContain("nemoclaw_e2e_configure_compatible_inference");
+  });
+
   it("keeps converted jobs dispatchable through the reusable workflow", () => {
     const cloudJob = nightlyWorkflow.jobs["cloud-e2e"];
 
diff --git a/test/e2e/test-token-rotation.sh b/test/e2e/test-token-rotation.sh
index f459be49c0..55be2ebf2f 100755
--- a/test/e2e/test-token-rotation.sh
+++ b/test/e2e/test-token-rotation.sh
@@ -16,7 +16,7 @@
 #
 # Prerequisites:
 #   - Docker running
-#   - NVIDIA_INFERENCE_API_KEY set (or fake OpenAI endpoint)
+#   - NVIDIA_INFERENCE_API_KEY set (or CI-compatible inference env)
 #   - TELEGRAM_BOT_TOKEN_A and TELEGRAM_BOT_TOKEN_B set (can be fake)
 #   - DISCORD_BOT_TOKEN_A and DISCORD_BOT_TOKEN_B set (can be fake)
 #   - SLACK_BOT_TOKEN_A and SLACK_BOT_TOKEN_B set (can be fake; xoxb- prefix)
@@ -37,6 +37,8 @@ export NEMOCLAW_E2E_DEFAULT_TIMEOUT=2400
 SCRIPT_DIR_TIMEOUT="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
 # shellcheck source=test/e2e/e2e-timeout.sh
 source "${SCRIPT_DIR_TIMEOUT}/e2e-timeout.sh"
+# shellcheck source=test/e2e/lib/ci-compatible-inference.sh
+. "${SCRIPT_DIR_TIMEOUT}/lib/ci-compatible-inference.sh"
 
 PASS=0
 FAIL=0
@@ -95,6 +97,8 @@ SANDBOX_NAME="${NEMOCLAW_SANDBOX_NAME:-e2e-token-rotation}"
 REGISTRY="$HOME/.nemoclaw/sandboxes.json"
 INSTALL_LOG="/tmp/nemoclaw-e2e-install.log"
 
+nemoclaw_e2e_configure_compatible_inference
+
 # ── Prerequisite checks ──────────────────────────────────────────
 
 if [ -z "${TELEGRAM_BOT_TOKEN_A:-}" ] || [ -z "${TELEGRAM_BOT_TOKEN_B:-}" ]; then