refactor(models): migrate models consumers to typed ModelsClient

e2e/agents_deploy_helpers.py

@@ -28,6 +28,8 @@
 import pytest
 from nemo_agents_plugin.entities import NAT_WORKFLOW_CONFIG_FORMAT, NEMO_AGENTS_SPEC_CONFIG_FORMAT
 from nemo_platform import NeMoPlatform
+from nemo_platform_plugin.client.adapter import client_from_platform
+from nemo_platform_plugin.models.client import ModelsClient
 from nmp.testing import MockProviderResponse, add_mock_provider
 
 # The mocked completion the deployed agent must round-trip back to the caller.
@@ -274,7 +276,7 @@ def run_agent_deploy_and_invoke(
             endpoints = deployment.get("endpoints") or []
             assert endpoints and endpoints[0]["url"], deployment
 
-        sdk.models.wait_for_openai_model(model_name, workspace=workspace)
+        client_from_platform(sdk, ModelsClient).wait_for_openai_model(model_name, workspace=workspace)
 
         response = sdk.agents.invoke(
             workspace=workspace,

e2e/test_evaluator_plugin.py

@@ -46,8 +46,12 @@
 from nemo_evaluator_sdk.metrics.tool_calling import ToolCallingMetric
 from nemo_evaluator_sdk.values.results import EvaluationResult
 from nemo_evaluator_sdk.values.scores import JSONScoreParser, RangeScore
-from nemo_platform import APIConnectionError, APIStatusError, NeMoPlatform
-from nemo_platform.types.inference import ModelProvider
+from nemo_platform import NeMoPlatform
+from nemo_platform_plugin.client.adapter import client_from_platform
+from nemo_platform_plugin.client.errors import NemoHTTPError as APIStatusError
+from nemo_platform_plugin.client.errors import NemoTransportError as APIConnectionError
+from nemo_platform_plugin.models.client import ModelsClient
+from nemo_platform_plugin.models.types import CreateModelEntityRequest, ModelProvider
 from nmp.testing import add_mock_provider, short_unique_name, wait_for_model_entity
 from nmp.testing.e2e import wait_for_platform_job
 from nmp.testing.utils import ensure_passthrough_virtual_model
@@ -261,13 +265,15 @@ def _create_ready_mock_model(
         mock_response_body=mock_response_body,
         should_autoprovision_virtual_model=False,
     )
-    sdk.models.create(
+    client_from_platform(sdk, ModelsClient).create_model(
         workspace=workspace,
-        name=name,
-        backend_format="OPENAI_CHAT",
-        model_providers=[f"{workspace}/{provider.name}"],
+        body=CreateModelEntityRequest(
+            name=name,
+            backend_format="OPENAI_CHAT",
+            model_providers=[f"{workspace}/{provider.name}"],
+        ),
         exist_ok=True,
-    )
+    ).data()
     wait_for_model_entity(
         sdk,
         workspace,

packages/nemo_nb/tests/test_myst_stripping.py

@@ -246,7 +246,7 @@ def test_mixed_content():
 
 ```python
 # Regular code
-sdk.models.deploy()
+sdk.models.create_deployment()
 ```
 
 :::{warning}
@@ -267,4 +267,4 @@ def test_mixed_content():
     assert "Use the CLI for faster deployment." in result
     assert "Make sure GPU resources are configured." in result
     assert "```python" in result
-    assert "sdk.models.deploy()" in result
+    assert "sdk.models.create_deployment()" in result

packages/nemo_nb/tests/test_notebook_splitting.py

@@ -399,7 +399,11 @@ def test_real_world_tab_set_example(self):
                 {"cell_type": "code", "metadata": {"language": "bash"}, "source": ["nemo models list\n"]},
                 {"cell_type": "markdown", "source": [":::\n"]},
                 {"cell_type": "markdown", "source": [":::{tab-item} Python SDK\n", ":sync: python-sdk\n"]},
-                {"cell_type": "code", "metadata": {"language": "python"}, "source": ["client.models.list()\n"]},
+                {
+                    "cell_type": "code",
+                    "metadata": {"language": "python"},
+                    "source": ["client_from_platform(client, ModelsClient).list_models()\n"],
+                },
                 {"cell_type": "markdown", "source": [":::\n"]},
                 {"cell_type": "markdown", "source": ["::::\n"]},
             ]

packages/nemo_nb/tests/test_strip_type_checker_comments.py

@@ -16,7 +16,7 @@ def test_strip_ty_ignore_comments():
                 "metadata": {"language": "python"},
                 "source": [
                     "# This is a regular comment\n",
-                    "sdk.models.get_openai_route_base_url()\n",
+                    "client_from_platform(sdk, ModelsClient).get_openai_route_base_url()\n",
                     "sdk.models.get_model_entity_route_openai_url(entity) # ty: ignore[unresolved-reference]\n",
                     "sdk.models.get_provider_route_openai_url(provider) # ty: ignore[unresolved-reference]\n",
                 ],
@@ -35,8 +35,8 @@ def test_strip_ty_ignore_comments():
     assert "# This is a regular comment" in result
 
     # Verify that the code lines are still there (without the ty: comments)
-    assert "sdk.models.get_model_entity_route_openai_url(entity)" in result
-    assert "sdk.models.get_provider_route_openai_url(provider)" in result
+    assert "client_from_platform(sdk, ModelsClient).get_model_entity_route_openai_url(entity)" in result
+    assert "client_from_platform(sdk, ModelsClient).get_provider_route_openai_url(provider)" in result
 
 
 def test_strip_type_ignore_comments():

packages/nemo_platform_ext/tests/cli/core/test_code_generator.py

@@ -18,7 +18,7 @@ def test_generate_python_code_simple_list():
 
     assert "from nemo_platform import NeMoPlatform" in code
     assert "client = NeMoPlatform()" in code
-    assert "response = client.models.list()" in code
+    assert "response = client_from_platform(client, ModelsClient).list_models()" in code
     assert "print(response)" in code
 
 

packages/nemo_platform_plugin/src/nemo_platform_plugin/agents/client.py

@@ -0,0 +1,33 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Typed HTTP clients for the Agents service.
+
+Wraps the endpoint functions from ``agents.endpoints`` as direct methods
+using the ``method()`` descriptor, following the files/models pattern.
+"""
+
+from nemo_platform_plugin.agents import endpoints
+from nemo_platform_plugin.client.client import AsyncNemoClient, NemoClient
+from nemo_platform_plugin.client.method import method
+
+
+class _AgentsMethods:
+    get_agent = method(endpoints.get_agent)
+    list_agents = method(endpoints.list_agents)
+    create_agent = method(endpoints.create_agent)
+    delete_agent = method(endpoints.delete_agent)
+    get_deployment = method(endpoints.get_deployment)
+    list_deployments = method(endpoints.list_deployments)
+    create_deployment = method(endpoints.create_deployment)
+    delete_deployment = method(endpoints.delete_deployment)
+    invoke_agent = method(endpoints.invoke_agent)
+    invoke_deployment = method(endpoints.invoke_deployment)
+
+
+class AgentsClient(_AgentsMethods, NemoClient):
+    """Sync client for the Agents service API."""
+
+
+class AsyncAgentsClient(_AgentsMethods, AsyncNemoClient):
+    """Async client for the Agents service API."""

packages/nemo_platform_plugin/src/nemo_platform_plugin/agents/endpoints.py

@@ -0,0 +1,102 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Typed endpoint definitions for the Agents service.
+
+Single source of truth for the HTTP contract. Replaces the Stainless-generated
+agent resource from ``nemo_agents_plugin.sdk``.
+"""
+
+from __future__ import annotations
+
+from abc import abstractmethod
+from typing import Any
+
+from nemo_platform_plugin.agents.types import (
+    Agent,
+    AgentDeployment,
+    CreateAgentDeploymentRequest,
+    CreateAgentRequest,
+    InvokeAgentRequest,
+    ListAgentsQueryParams,
+    ListDeploymentsQueryParams,
+)
+from nemo_platform_plugin.client.endpoint import delete, get, post
+from nemo_platform_plugin.client.types import Paginated, PreparedRequest
+
+_AGENTS = "/apis/agents/v2/workspaces/{workspace}/agents"
+_DEPLOYMENTS = "/apis/agents/v2/workspaces/{workspace}/deployments"
+
+
+# ---------------------------------------------------------------------------
+# Agent CRUD
+# ---------------------------------------------------------------------------
+
+
+@get(f"{_AGENTS}/{{name}}")
+@abstractmethod
+def get_agent(*, workspace: str | None = None, name: str) -> Agent: ...
+
+
+@get(_AGENTS)
+@abstractmethod
+def list_agents(
+    *, workspace: str | None = None, query_params: ListAgentsQueryParams | None = None
+) -> Paginated[Agent]: ...
+
+
+def _get_agent_on_conflict(body: CreateAgentRequest, workspace: str | None) -> PreparedRequest[Agent]:
+    """Build the retrieve request replayed when ``create_agent(exist_ok=True)`` 409s."""
+    return get_agent(name=body.name, workspace=workspace)
+
+
+@post(_AGENTS, get_on_conflict=_get_agent_on_conflict)
+@abstractmethod
+def create_agent(*, workspace: str | None = None, body: CreateAgentRequest, exist_ok: bool = False) -> Agent: ...
+
+
+@delete(f"{_AGENTS}/{{name}}")
+@abstractmethod
+def delete_agent(*, workspace: str | None = None, name: str) -> None: ...
+
+
+# ---------------------------------------------------------------------------
+# Agent deployment CRUD
+# ---------------------------------------------------------------------------
+
+
+@get(f"{_DEPLOYMENTS}/{{name}}")
+@abstractmethod
+def get_deployment(*, workspace: str | None = None, name: str) -> AgentDeployment: ...
+
+
+@get(_DEPLOYMENTS)
+@abstractmethod
+def list_deployments(
+    *, workspace: str | None = None, query_params: ListDeploymentsQueryParams | None = None
+) -> Paginated[AgentDeployment]: ...
+
+
+@post(_DEPLOYMENTS)
+@abstractmethod
+def create_deployment(*, workspace: str | None = None, body: CreateAgentDeploymentRequest) -> AgentDeployment: ...
+
+
+@delete(f"{_DEPLOYMENTS}/{{name}}")
+@abstractmethod
+def delete_deployment(*, workspace: str | None = None, name: str) -> None: ...
+
+
+# ---------------------------------------------------------------------------
+# Agent invocation (gateway proxy to OpenAI-compatible endpoint)
+# ---------------------------------------------------------------------------
+
+
+@post(f"{_AGENTS}/{{name}}/-/v1/chat/completions")
+@abstractmethod
+def invoke_agent(*, workspace: str | None = None, name: str, body: InvokeAgentRequest) -> dict[str, Any]: ...
+
+
+@post(f"{_DEPLOYMENTS}/{{name}}/-/v1/chat/completions")
+@abstractmethod
+def invoke_deployment(*, workspace: str | None = None, name: str, body: InvokeAgentRequest) -> dict[str, Any]: ...

packages/nemo_platform_plugin/src/nemo_platform_plugin/agents/types.py

@@ -0,0 +1,107 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Shared request/response types for the Agents service.
+
+Single source of truth for the HTTP contract. Replaces the Stainless-generated
+agent resource from ``nemo_agents_plugin.sdk``.
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Any, NotRequired, TypedDict
+
+from nemo_platform_plugin.schema import Page
+from pydantic import BaseModel, ConfigDict, Field
+
+# ---------------------------------------------------------------------------
+# Response types
+# ---------------------------------------------------------------------------
+
+
+class Agent(BaseModel):
+    """An agent definition — stores agent config and metadata."""
+
+    model_config = ConfigDict(extra="allow")
+
+    name: str = ""
+    workspace: str = ""
+    project: str | None = None
+    description: str = ""
+    config: dict[str, Any] = Field(default_factory=dict)
+    config_format: str = "nat-workflow-v1"
+    id: str | None = None
+    created_at: datetime | None = None
+    created_by: str | None = None
+    updated_at: datetime | None = None
+    updated_by: str | None = None
+
+
+AgentPage = Page[Agent]
+
+
+class AgentDeployment(BaseModel):
+    """A running (or pending) deployment of an Agent."""
+
+    model_config = ConfigDict(extra="allow")
+
+    name: str = ""
+    workspace: str = ""
+    project: str | None = None
+    agent: str = ""
+    config: dict[str, Any] = Field(default_factory=dict)
+    status: str = "pending"
+    deployment_mode: str = "subprocess"
+    endpoint: str = ""
+    id: str | None = None
+    created_at: datetime | None = None
+    updated_at: datetime | None = None
+
+
+DeploymentPage = Page[AgentDeployment]
+
+
+# ---------------------------------------------------------------------------
+# Request types
+# ---------------------------------------------------------------------------
+
+
+class CreateAgentRequest(BaseModel):
+    name: str
+    description: str = ""
+    config: dict[str, Any] = Field(default_factory=dict)
+    config_format: str = "nat-workflow-v1"
+
+
+class CreateAgentDeploymentRequest(BaseModel):
+    agent: str
+    config: dict[str, Any] = Field(default_factory=dict)
+
+
+class InvokeAgentRequest(BaseModel):
+    """OpenAI chat-completions request body for agent invocation."""
+
+    model_config = ConfigDict(extra="allow")
+
+    messages: list[dict[str, Any]] = Field(default_factory=list)
+    stream: bool = False
+
+
+# ---------------------------------------------------------------------------
+# Query parameter types
+# ---------------------------------------------------------------------------
+
+
+class ListAgentsQueryParams(TypedDict, total=False):
+    page: NotRequired[int]
+    page_size: NotRequired[int]
+    sort: NotRequired[str]
+    filter: NotRequired[str]
+
+
+class ListDeploymentsQueryParams(TypedDict, total=False):
+    page: NotRequired[int]
+    page_size: NotRequired[int]
+    sort: NotRequired[str]
+    filter: NotRequired[str]

packages/nemo_platform_plugin/src/nemo_platform_plugin/auditor/client.py

@@ -0,0 +1,36 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Typed HTTP clients for the Auditor service.
+
+Wraps the endpoint functions from ``auditor.endpoints`` as direct methods
+using the ``method()`` descriptor, following the files/models pattern.
+"""
+
+from nemo_platform_plugin.auditor import endpoints
+from nemo_platform_plugin.client.client import AsyncNemoClient, NemoClient
+from nemo_platform_plugin.client.method import method
+
+
+class _AuditorMethods:
+    get_audit_config = method(endpoints.get_audit_config)
+    list_audit_configs = method(endpoints.list_audit_configs)
+    create_audit_config = method(endpoints.create_audit_config)
+    update_audit_config = method(endpoints.update_audit_config)
+    delete_audit_config = method(endpoints.delete_audit_config)
+    get_audit_target = method(endpoints.get_audit_target)
+    list_audit_targets = method(endpoints.list_audit_targets)
+    create_audit_target = method(endpoints.create_audit_target)
+    update_audit_target = method(endpoints.update_audit_target)
+    delete_audit_target = method(endpoints.delete_audit_target)
+    submit_audit = method(endpoints.submit_audit)
+    list_audit_jobs = method(endpoints.list_audit_jobs)
+    get_audit_job = method(endpoints.get_audit_job)
+
+
+class AuditorClient(_AuditorMethods, NemoClient):
+    """Sync client for the Auditor service API."""
+
+
+class AsyncAuditorClient(_AuditorMethods, AsyncNemoClient):
+    """Async client for the Auditor service API."""