fix(agents): Honor shared CLI context for base URL and auth

plugins/nemo-agents/src/nemo_agents_plugin/cli.py

@@ -45,6 +45,18 @@
 import httpx
 import typer
 import yaml
+from nemo_agents_plugin.cli_context import (
+    DEFAULT_BASE_URL as _DEFAULT_BASE_URL,
+)
+from nemo_agents_plugin.cli_context import (
+    BaseUrlOption,
+)
+from nemo_agents_plugin.cli_context import (
+    resolve_base_url as _resolve_base_url,
+)
+from nemo_agents_plugin.cli_context import (
+    resolve_context_headers as _resolve_context_headers,
+)
 from nemo_agents_plugin.leaderboard.cli import register_leaderboard_commands
 from nemo_agents_plugin.usage.cli import register_usage_commands
 from nemo_platform.cli.core.formatters import Column, format_output
@@ -54,7 +66,6 @@
 
 logger = logging.getLogger(__name__)
 
-_DEFAULT_BASE_URL = "http://localhost:8080"
 _DEFAULT_WORKSPACE = "default"
 _LIST_OUTPUT_FORMAT = Literal["table", "json", "yaml", "csv", "markdown", "raw"]
 _AGENT_LIST_COLUMNS = [
@@ -145,7 +156,7 @@ def invoke(
             help="Name of a specific deployment to invoke (platform required).",
         ),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         timeout: float = typer.Option(
             300,
             "--timeout",
@@ -160,6 +171,7 @@ def invoke(
         ),
     ) -> None:
         """Invoke an agent — locally (with --agent-config) or via the platform (with --agent or --agent-deployment)."""
+        base_url = _resolve_base_url(base_url)
         if agent_config:
             _local_invoke(agent_config, input, input_file, workspace=workspace, base_url=base_url)
         elif agent or agent_deployment:
@@ -628,9 +640,10 @@ def create(
         ),
         description: str = typer.Option("", "--description"),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Register an agent on the platform."""
+        base_url = _resolve_base_url(base_url)
         from nemo_agents_plugin.utils import inject_default_model
 
         config_dict = _load_yaml(agent_config)
@@ -653,7 +666,7 @@ def create(
     def list_agents(
         ctx: typer.Context,
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         output_format: Optional[_LIST_OUTPUT_FORMAT] = typer.Option(
             None,
             "--format",
@@ -671,6 +684,7 @@ def list_agents(
         ),
     ) -> None:
         """List agents on the platform."""
+        base_url = _resolve_base_url(base_url)
         resp = _api_request("GET", base_url, f"/apis/agents/v2/workspaces/{workspace}/agents")
         _print_list_response(
             ctx,
@@ -684,20 +698,22 @@ def list_agents(
     def get(
         name: str = typer.Argument(..., help="Agent name."),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Get an agent by name."""
+        base_url = _resolve_base_url(base_url)
         resp = _api_request("GET", base_url, f"/apis/agents/v2/workspaces/{workspace}/agents/{name}")
         typer.echo(json.dumps(resp, indent=2))
 
     @app.command(rich_help_panel="Agent Resources (requires running cluster)")
     def delete(
         name: str = typer.Argument(..., help="Agent name."),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         yes: bool = typer.Option(False, "--yes", "-y", help="Skip confirmation prompt."),
     ) -> None:
         """Delete an agent from the platform."""
+        base_url = _resolve_base_url(base_url)
         if not yes:
             typer.confirm(f"Delete agent '{name}'?", abort=True)
         _api_request("DELETE", base_url, f"/apis/agents/v2/workspaces/{workspace}/agents/{name}")
@@ -725,7 +741,7 @@ def deploy(
             help="Maximum seconds to wait for a terminal status (only with --wait).",
         ),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Deploy an agent on the platform.
 
@@ -736,6 +752,7 @@ def deploy(
         scripted pipelines that prefer to poll separately via ``nemo agents
         deployments wait``.
         """
+        base_url = _resolve_base_url(base_url)
         payload: dict = {"agent": agent}
         if name:
             payload["name"] = name
@@ -792,7 +809,7 @@ def logs(
             help="Print only the absolute log file path and exit (useful for scripting).",
         ),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Show logs for an agent deployment.
 
@@ -816,6 +833,7 @@ def logs(
             raise typer.Exit(code=1)
 
         if agent and not name:
+            base_url = _resolve_base_url(base_url)
             candidates = [
                 d
                 for d in _unwrap_list(
@@ -859,10 +877,11 @@ def undeploy(
             None, "--agent", "--all", "-a", help="Remove all deployments for this agent."
         ),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         yes: bool = typer.Option(False, "--yes", "-y", help="Skip confirmation prompt."),
     ) -> None:
         """Stop and remove a deployment (or all deployments for an agent)."""
+        base_url = _resolve_base_url(base_url)
         if name:
             if not yes:
                 typer.confirm(f"Undeploy '{name}'?", abort=True)
@@ -888,7 +907,7 @@ def undeploy(
     def deployments_list(
         ctx: typer.Context,
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         output_format: Optional[_LIST_OUTPUT_FORMAT] = typer.Option(
             None,
             "--format",
@@ -906,6 +925,7 @@ def deployments_list(
         ),
     ) -> None:
         """List deployments."""
+        base_url = _resolve_base_url(base_url)
         resp = _api_request("GET", base_url, f"/apis/agents/v2/workspaces/{workspace}/deployments")
         _print_list_response(
             ctx,
@@ -919,20 +939,22 @@ def deployments_list(
     def deployments_get(
         name: str = typer.Argument(..., help="Deployment name."),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Get a deployment by name."""
+        base_url = _resolve_base_url(base_url)
         resp = _api_request("GET", base_url, f"/apis/agents/v2/workspaces/{workspace}/deployments/{name}")
         typer.echo(json.dumps(resp, indent=2))
 
     @deps_app.command(name="delete")
     def deployments_delete(
         name: str = typer.Argument(..., help="Deployment name."),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
         yes: bool = typer.Option(False, "--yes", "-y", help="Skip confirmation prompt."),
     ) -> None:
         """Delete a deployment by name."""
+        base_url = _resolve_base_url(base_url)
         if not yes:
             typer.confirm(f"Delete deployment '{name}'?", abort=True)
         _api_request("DELETE", base_url, f"/apis/agents/v2/workspaces/{workspace}/deployments/{name}")
@@ -950,7 +972,7 @@ def deployments_wait(
         timeout: int = typer.Option(300, "--timeout", "-t", help="Maximum seconds to wait."),
         interval: float = typer.Option(2.0, "--interval", help="Poll interval in seconds."),
         workspace: str = typer.Option(_DEFAULT_WORKSPACE, "--workspace", "-w"),
-        base_url: str = typer.Option(_DEFAULT_BASE_URL, "--base-url", envvar="NEMO_BASE_URL"),
+        base_url: BaseUrlOption = None,
     ) -> None:
         """Wait for a deployment to reach 'running' or 'failed' status.
 
@@ -960,6 +982,7 @@ def deployments_wait(
         Provide either a deployment name directly or --agent to resolve the
         latest active deployment for that agent automatically.
         """
+        base_url = _resolve_base_url(base_url)
         if not name and not agent:
             typer.echo("Error: provide a deployment name or --agent.", err=True)
             raise typer.Exit(code=1)
@@ -1200,13 +1223,14 @@ def _platform_invoke(
         path = f"/apis/agents/v2/workspaces/{workspace}/deployments/{deployment}/-/v1/chat/completions"
 
     url = base_url.rstrip("/") + path
+    headers = _resolve_context_headers()
     target_label = agent or deployment
     for query in queries:
         payload = {"messages": [{"role": "user", "content": query}], "stream": False}
         try:
             with request_progress(f"Waiting for agent '{target_label}'...", disabled=no_progress):
                 with httpx.Client(timeout=timeout) as client:
-                    resp = client.post(url, json=payload)
+                    resp = client.post(url, json=payload, headers=headers or None)
                     resp.raise_for_status()
                     body = resp.json()
                 typer.echo(json.dumps(body, indent=2))
@@ -1299,6 +1323,9 @@ def _api_request(method: str, base_url: str, path: str, *, json_body: dict[str,
     request_kwargs: dict[str, Any] = {}
     if json_body is not None:
         request_kwargs["json"] = json_body
+    headers = _resolve_context_headers()
+    if headers:
+        request_kwargs["headers"] = headers
     try:
         with httpx.Client(timeout=30) as client:
             resp = client.request(method, url, **request_kwargs)

plugins/nemo-agents/src/nemo_agents_plugin/cli_context.py

@@ -0,0 +1,111 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Shared CLI-context resolution for the ``nemo agents`` command group.
+
+The agents plugin makes platform calls from two places — the platform
+commands in :mod:`nemo_agents_plugin.cli` (raw ``httpx``) and
+``nemo agents usage show`` in :mod:`nemo_agents_plugin.usage.cli` (the
+NeMoPlatform SDK client).  Both must resolve the platform base URL and the
+auth token the same way every other ``nemo`` command does: through the
+shared CLI context object stored on ``typer.Context.obj``.
+
+These helpers read the *ambient* Click context so callers deep in a command's
+call stack can resolve configuration without threading the context object
+through every function signature.  They live in their own module (rather than
+in ``cli.py``) because ``cli.py`` imports the usage CLI at module load, so a
+back-import would create a cycle.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Annotated, Any, Optional
+
+import click
+import typer
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_BASE_URL = "http://localhost:8080"
+
+BASE_URL_HELP = (
+    "Platform base URL. Resolution order: "
+    "(1) this --base-url flag or NEMO_BASE_URL; "
+    "(2) shared CLI config (`nemo config set --base-url`) or NMP_BASE_URL; "
+    f"(3) {DEFAULT_BASE_URL} (default)."
+)
+
+# Reusable ``--base-url`` option shared across every ``nemo agents`` command
+# so the option and its help text are defined once. ``None`` means "unset" so
+# ``resolve_base_url`` can fall back to the shared CLI context / config.
+BaseUrlOption = Annotated[
+    Optional[str],
+    typer.Option("--base-url", envvar="NEMO_BASE_URL", help=BASE_URL_HELP),
+]
+
+
+def current_cli_state() -> Any:
+    """Return the shared CLI context object (``typer.Context.obj``) if present.
+
+    Returns ``None`` when the plugin is exercised outside a Click invocation
+    (e.g. a direct unit test), so callers fall back to their own defaults.
+    """
+    ctx = click.get_current_context(silent=True)
+    return ctx.obj if ctx is not None else None
+
+
+def base_url_from_context() -> str | None:
+    """Return the base URL configured in the shared CLI context, if any."""
+    state = current_cli_state()
+    if state is None or not hasattr(state, "get_base_url"):
+        return None
+    try:
+        return state.get_base_url(default=None)
+    except Exception:
+        logger.debug("Failed to resolve base URL from CLI context", exc_info=True)
+        return None
+
+
+def resolve_base_url(base_url: str | None) -> str:
+    """Resolve the platform base URL and announce the target on stderr.
+
+    Precedence:
+      1. Explicit ``--base-url`` / ``NEMO_BASE_URL`` on the command.
+      2. The shared CLI context — ``nemo config set --base-url`` and the
+         ``NMP_BASE_URL`` env var — so ``nemo agents`` targets the same
+         platform as every other ``nemo`` command.
+      3. The built-in localhost default.
+
+    The resolved target is echoed to stderr (never stdout, so piped/JSON
+    output stays clean) so a mis-pointed command is visible instead of
+    silently hitting the wrong platform.
+    """
+    resolved = base_url or base_url_from_context() or DEFAULT_BASE_URL
+    click.echo(f"Targeting {resolved}", err=True)
+    return resolved
+
+
+def resolve_context_headers() -> dict[str, str]:
+    """Return auth (and other) default headers from the shared CLI context.
+
+    Mirrors ``nemo_platform_plugin.commands._resolve_submit_auth_headers``:
+    reads the SDK client config off the shared context so ``nemo agents``
+    attaches the same ``Authorization: Bearer`` token as the rest of the CLI
+    (i.e. the token established by ``nemo auth login``).  Returns an empty
+    mapping when no context or token is available — leaving requests
+    unauthenticated exactly as before, so local unauthenticated dev keeps
+    working.
+    """
+    state = current_cli_state()
+    if state is None or not hasattr(state, "get_sdk_context"):
+        return {}
+    try:
+        client_config = state.get_sdk_context().user.get_client_config()
+    except Exception:
+        logger.debug("Failed to resolve auth headers from CLI context", exc_info=True)
+        return {}
+    headers = client_config.get("default_headers") if isinstance(client_config, dict) else None
+    if isinstance(headers, dict):
+        return {str(key): str(value) for key, value in headers.items()}
+    return {}