fix(cli): accept active sessions autoscaling metric

src/together/lib/cli/api/beta/endpoints/_utils/_build_autoscaling.py

@@ -19,6 +19,7 @@
 ]
 
 ScalingMetricName = Literal[
+    "active_sessions",
     "inflight_requests",
     "gpu_utilization",
     "token_utilization",
@@ -33,6 +34,7 @@
 
 # Fixed type per metric name (see examples/internal-team-guides/autoscaling.md).
 _METRIC_TYPES: dict[ScalingMetricName, MetricType] = {
+    "active_sessions": "METRIC_TARGET_TYPE_VALUE",
     "inflight_requests": "METRIC_TARGET_TYPE_AVERAGE_VALUE",
     "gpu_utilization": "METRIC_TARGET_TYPE_UTILIZATION",
     "token_utilization": "METRIC_TARGET_TYPE_UTILIZATION",

src/together/lib/cli/api/beta/endpoints/deploy.py

@@ -125,6 +125,7 @@ async def deploy(
             help=(
                 """Autoscaling metric. Must be set with --scaling-target; --scaling-percentile is optional and only applies to latency metrics.
 
+- active_sessions: Active sessions across the deployment.
 - inflight_requests: Concurrent in-flight requests per replica.
 - gpu_utilization: GPU compute utilization (%).
 - token_utilization: KV-cache utilization (%).

tests/cli/test_beta_endpoints.py

@@ -377,6 +377,47 @@ def test_deploy_onto_existing_endpoint(self, respx_mock: MockRouter, cli_runner:
         assert result.exit_code == 0, result.output
         assert create_deployment_route.call_count == 1
 
+    @pytest.mark.respx(base_url=base_url)
+    def test_deploy_accepts_active_sessions_scaling_metric(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        _mock_model_and_config(respx_mock)
+        respx_mock.get("/projects/proj/endpoints/ep_1").mock(return_value=httpx.Response(200, json=_endpoint_body()))
+        create_deployment_route = respx_mock.post("/projects/proj/endpoints/ep_1/deployments").mock(
+            return_value=httpx.Response(200, json=_deployment_body())
+        )
+
+        result = cli_runner.invoke(
+            [
+                "beta",
+                "endpoints",
+                "deploy",
+                "--project",
+                "proj",
+                "--endpoint",
+                "ep_1",
+                "--model",
+                "ml_1",
+                "--config",
+                "cr_1",
+                "--deployment-name",
+                "my-dep",
+                "--scaling-metric",
+                "active_sessions",
+                "--scaling-target",
+                "25",
+                "--json",
+            ]
+        )
+
+        assert result.exit_code == 0, result.output
+        deployment_body = json.loads(cast(Call, create_deployment_route.calls[0]).request.content.decode())
+        assert deployment_body["autoscaling"]["scalingMetrics"] == [
+            {
+                "name": "active_sessions",
+                "type": "METRIC_TARGET_TYPE_VALUE",
+                "target": 25.0,
+            }
+        ]
+
     @pytest.mark.respx(base_url=base_url)
     def test_deploy_reuses_endpoint_when_name_already_exists(
         self, respx_mock: MockRouter, cli_runner: CliRunner

tests/cli/test_beta_endpoints_update.py

@@ -169,6 +169,32 @@ def test_update_by_deployment_id(self, respx_mock: MockRouter, cli_runner: CliRu
             "autoscaling": {"minReplicas": 1, "maxReplicas": 2},
         }
 
+    @pytest.mark.respx(base_url=base_url)
+    def test_update_accepts_active_sessions_scaling_metric(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        _mock_endpoint_list(respx_mock)
+        route = respx_mock.patch("/projects/proj/endpoints/ep_1/deployments/dep_control").mock(
+            return_value=httpx.Response(200, json=_deployment_body())
+        )
+
+        result = cli_runner.invoke(
+            _update_args("dep_control", "--scaling-metric", "active_sessions", "--scaling-target", "25")
+        )
+
+        assert result.exit_code == 0, result.output
+        req = cast(Call, route.calls[0]).request
+        assert "updateMask=autoscaling" in str(req.url)
+        assert json.loads(req.content.decode()) == {
+            "autoscaling": {
+                "scalingMetrics": [
+                    {
+                        "name": "active_sessions",
+                        "type": "METRIC_TARGET_TYPE_VALUE",
+                        "target": 25.0,
+                    }
+                ]
+            },
+        }
+
     @pytest.mark.respx(base_url=base_url)
     def test_update_idle_deployment(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
         _mock_endpoint_list(respx_mock)

tests/cli/test_build_autoscaling.py

@@ -18,6 +18,16 @@ def test_build_scaling_metrics_utilization() -> None:
     ]
 
 
+def test_build_scaling_metrics_active_sessions() -> None:
+    assert build_scaling_metrics(scaling_metric="active_sessions", scaling_target=25) == [
+        {
+            "name": "active_sessions",
+            "type": "METRIC_TARGET_TYPE_VALUE",
+            "target": 25,
+        }
+    ]
+
+
 def test_build_scaling_metrics_latency_with_percentile() -> None:
     assert build_scaling_metrics(
         scaling_metric="ttft",