feat: add nemotron to tool_preset options in CI workflows

Add nemotron to:
- ToolPresetType literal in tests/integration/base.py
- get_tools_for_preset() function to return nemotron tools
- run-eval.yml workflow tool_preset dropdown
- integration-runner.yml workflow tool_preset dropdown
- run_infer.py argparse choices
- test_tool_presets.py for nemotron validation

This enables running evaluations with TOOL_PRESET=nemotron to test the
Nemotron-3 Super model with its native tool names (bash, str_replace).

Co-authored-by: openhands <openhands@all-hands.dev>

Author: openhands-agent

.github/workflows/integration-runner.yml

@@ -33,9 +33,9 @@ on:
                 type: string
             tool_preset:
                 description: >-
-                    Tool preset for file editing (default, gemini, gpt5, planning).
+                    Tool preset for file editing (default, gemini, gpt5, planning, nemotron).
                     'default' uses FileEditorTool, 'gemini' uses read_file/write_file/edit/list_directory,
-                    'gpt5' uses apply_patch tool.
+                    'gpt5' uses apply_patch tool, 'nemotron' uses bash/str_replace (Anthropic-compatible).
                 required: false
                 default: default
                 type: choice
@@ -44,6 +44,7 @@ on:
                     - gemini
                     - gpt5
                     - planning
+                    - nemotron
     schedule:
         - cron: 30 22 * * * # Runs at 10:30pm UTC every day
 

.github/workflows/run-eval.yml

@@ -92,7 +92,8 @@ on:
                 description: >-
                     Tool preset for file editing. 'default' uses FileEditorTool,
                     'gemini' uses read_file/write_file/edit/list_directory,
-                    'gpt5' uses apply_patch tool.
+                    'gpt5' uses apply_patch tool,
+                    'nemotron' uses bash/str_replace (Anthropic-compatible).
                 required: false
                 default: default
                 type: choice
@@ -101,6 +102,7 @@ on:
                     - gemini
                     - gpt5
                     - planning
+                    - nemotron
             agent_type:
                 description: >-
                     Agent type: 'default' for standard Agent,

tests/integration/base.py

@@ -29,7 +29,7 @@
 
 
 # Tool preset type for selecting which file editing toolset to use
-ToolPresetType = Literal["default", "gemini", "gpt5", "planning"]
+ToolPresetType = Literal["default", "gemini", "gpt5", "planning", "nemotron"]
 
 
 def get_tools_for_preset(
@@ -38,7 +38,7 @@ def get_tools_for_preset(
     """Get the list of tools for the given preset.
 
     Args:
-        preset: The tool preset to use (default, gemini, gpt5, or planning).
+        preset: The tool preset to use (default, gemini, gpt5, planning, or nemotron).
         enable_browser: Whether to include browser tools.
 
     Returns:
@@ -58,6 +58,10 @@ def get_tools_for_preset(
 
             # Planning preset is read-only and doesn't support browser tools
             return get_planning_tools()
+        case "nemotron":
+            from openhands.tools.preset.nemotron import get_nemotron_tools
+
+            return get_nemotron_tools(enable_browser=enable_browser)
         case "default":
             from openhands.tools.preset.default import get_default_tools
 

tests/integration/run_infer.py

@@ -455,14 +455,15 @@ def main():
     parser.add_argument(
         "--tool-preset",
         type=str,
-        choices=["default", "gemini", "gpt5", "planning"],
+        choices=["default", "gemini", "gpt5", "planning", "nemotron"],
         default="default",
         help=(
             "Tool preset to use for file editing (default: 'default'). "
             "'default' uses FileEditorTool (claude-style), "
             "'gemini' uses read_file/write_file/edit/list_directory tools, "
             "'gpt5' uses apply_patch tool, "
-            "'planning' uses planning-specific tools."
+            "'planning' uses planning-specific tools, "
+            "'nemotron' uses bash/str_replace tools (Anthropic-compatible)."
         ),
     )
 

tests/integration/test_tool_presets.py

@@ -80,10 +80,29 @@ def test_get_tools_for_preset_invalid():
         get_tools_for_preset("invalid_preset", enable_browser=False)  # type: ignore[arg-type]
 
 
+def test_get_tools_for_preset_nemotron():
+    """Test that nemotron preset returns Anthropic-compatible tools."""
+    tools = get_tools_for_preset("nemotron", enable_browser=False)
+    tool_names = {t.name for t in tools}
+
+    assert "bash" in tool_names
+    assert "str_replace" in tool_names
+    assert "task_tracker" in tool_names
+    # Default terminal/file_editor should NOT be present
+    assert "terminal" not in tool_names
+    assert "file_editor" not in tool_names
+
+
 def test_tool_preset_type_literal_values():
     """Verify ToolPresetType includes all expected values."""
     # This is a compile-time check but we document expected values here
-    valid_presets: list[ToolPresetType] = ["default", "gemini", "gpt5", "planning"]
+    valid_presets: list[ToolPresetType] = [
+        "default",
+        "gemini",
+        "gpt5",
+        "planning",
+        "nemotron",
+    ]
     for preset in valid_presets:
         # Should not raise
         tools = get_tools_for_preset(preset, enable_browser=False)
@@ -104,12 +123,18 @@ def test_run_infer_argparse_accepts_all_tool_presets():
     parser.add_argument(
         "--tool-preset",
         type=str,
-        choices=["default", "gemini", "gpt5", "planning"],
+        choices=["default", "gemini", "gpt5", "planning", "nemotron"],
         default="default",
     )
 
     # Test each valid preset value
-    valid_presets: list[ToolPresetType] = ["default", "gemini", "gpt5", "planning"]
+    valid_presets: list[ToolPresetType] = [
+        "default",
+        "gemini",
+        "gpt5",
+        "planning",
+        "nemotron",
+    ]
 
     for preset in valid_presets:
         # This should not raise an error