feat(nodes): add NVIDIA Nemotron LLM node

.github/workflows/sync-models.yml

@@ -63,6 +63,7 @@ jobs:
             [llm_qwen]=ROCKETRIDE_QWEN_KEY
             [llm_minimax]=ROCKETRIDE_MINIMAX_KEY
             [llm_kimi]=ROCKETRIDE_KIMI_KEY
+            [llm_nemotron]=ROCKETRIDE_NVIDIA_KEY
           )
 
           for PROVIDER in "${!PROVIDER_KEY[@]}"; do
@@ -90,6 +91,7 @@ jobs:
           ROCKETRIDE_QWEN_KEY: ${{ secrets.ROCKETRIDE_QWEN_KEY }}
           ROCKETRIDE_MINIMAX_KEY: ${{ secrets.ROCKETRIDE_MINIMAX_KEY }}
           ROCKETRIDE_KIMI_KEY: ${{ secrets.ROCKETRIDE_KIMI_KEY }}
+          ROCKETRIDE_NVIDIA_KEY: ${{ secrets.ROCKETRIDE_NVIDIA_KEY }}
 
       # Strict sync: only providers whose API key is configured. Native API
       # discovers and smoke-tests new models. No OpenRouter pollution.
@@ -110,6 +112,7 @@ jobs:
           ROCKETRIDE_QWEN_KEY: ${{ secrets.ROCKETRIDE_QWEN_KEY }}
           ROCKETRIDE_MINIMAX_KEY: ${{ secrets.ROCKETRIDE_MINIMAX_KEY }}
           ROCKETRIDE_KIMI_KEY: ${{ secrets.ROCKETRIDE_KIMI_KEY }}
+          ROCKETRIDE_NVIDIA_KEY: ${{ secrets.ROCKETRIDE_NVIDIA_KEY }}
 
       # Fallback sync: providers without an API key. OpenRouter / LiteLLM
       # are permitted as discovery sources for these. Smoke tests are skipped

docs/README-nodes.md

@@ -81,6 +81,7 @@ the same `tool_github` or `tool_tavily` can attach to `agent_deepagent`,
 | `llm_perplexity` | questions → answers  | Perplexity Sonar (web-search models)         |
 | `llm_gmi_cloud`  | questions → answers  | GMI Cloud models                             |
 | `llm_baidu_qianfan` | questions → answers | Baidu Qianfan / ERNIE (OpenAI-compatible Qianfan API) |
+| `llm_nemotron`   | questions → answers  | NVIDIA Nemotron reasoning models (build.nvidia.com API or self-hosted NIM) |
 | `llm_openai_api` | questions → answers  | Any OpenAI-compatible endpoint (also ships a Nebius Token Factory preset) |
 
 > `nodes/src/nodes/llm_ibm_watson/` exists but currently ships **no service

nodes/src/nodes/llm_nemotron/IGlobal.py

@@ -0,0 +1,151 @@
+# =============================================================================
+# MIT License
+# Copyright (c) 2026 Aparavi Software AG
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+# =============================================================================
+
+import os
+import re
+from typing import Optional
+from rocketlib import IGlobalBase, warning
+from ai.common.config import Config
+from ai.common.chat import ChatBase
+
+
+VALIDATION_PROMPT = 'Hi'
+NVIDIA_BASE_URL = 'https://integrate.api.nvidia.com/v1'
+
+
+class IGlobal(IGlobalBase):
+    """Global handler for the NVIDIA Nemotron LLM node."""
+
+    chat: Optional[ChatBase] = None
+
+    def validateConfig(self):
+        """Validate only cloud Nemotron models (NVIDIA API) at save time.
+
+        Local/self-hosted OpenAI-compatible endpoints (NIM / vLLM / SGLang)
+        are not validated here, matching the llm_kimi precedent.
+        """
+        # Load dependencies first
+        from depends import depends  # type: ignore
+
+        requirements = os.path.dirname(os.path.realpath(__file__)) + '/requirements.txt'
+        depends(requirements)
+
+        try:
+            # Import OpenAI SDK (the NVIDIA API is OpenAI-compatible)
+            from openai import OpenAI
+
+            # Prefer provider-driven exception types over string parsing
+            from openai import APIStatusError, OpenAIError, AuthenticationError, RateLimitError, APIConnectionError
+
+            # Get config
+            config = Config.getNodeConfig(self.glb.logicalType, self.glb.connConfig)
+            apikey = config.get('apikey')
+            model = config.get('model')
+            serverbase = config.get('serverbase') or NVIDIA_BASE_URL
+
+            # Only validate cloud NVIDIA endpoints; skip self-hosted/offline ones.
+            if 'api.nvidia.com' not in serverbase:
+                return
+
+            # UI handles missing key prompts; skip the probe when the key is absent
+            if not apikey or not model:
+                return
+
+            # Minimal probe; UI handles missing key prompts
+            try:
+                # Create client and make a 1-token probe
+                client = OpenAI(api_key=apikey, base_url=serverbase)
+                client.chat.completions.create(
+                    model=model, messages=[{'role': 'user', 'content': VALIDATION_PROMPT}], max_tokens=1
+                )
+            except APIStatusError as e:
+                # HTTP error with structured body; pull code/type/message
+                status = getattr(e, 'status_code', None) or getattr(e, 'status', None)
+                try:
+                    resp = getattr(e, 'response', None)
+                    data = resp.json() if resp is not None else None
+                    if isinstance(data, dict):
+                        err = data.get('error')
+                        if isinstance(err, dict):
+                            etype = err.get('type')
+                            emsg = err.get('message') or data.get('message')
+                        else:
+                            etype = None
+                            emsg = data.get('message')
+                        message = self._format_error(status, etype, emsg, str(e))
+                    else:
+                        message = self._format_error(status, None, None, str(e))
+                except Exception:
+                    message = self._format_error(status, None, None, str(e))
+                warning(message)
+                return
+            except (AuthenticationError, RateLimitError, APIConnectionError, OpenAIError) as e:
+                # Other OpenAI exceptions; format consistently
+                message = self._format_error(None, None, None, str(e))
+                warning(message)
+                return
+
+        except Exception as e:
+            # Generic fallback - do not alter provider message
+            warning(str(e))
+
+    def beginGlobal(self):
+        from depends import depends  # type: ignore
+
+        # Load the requirements
+        requirements = os.path.dirname(os.path.realpath(__file__)) + '/requirements.txt'
+        depends(requirements)
+
+        from .nemotron import Chat
+
+        # Get our bag
+        bag = self.IEndpoint.endpoint.bag
+
+        # Get this nodes config
+        config = Config.getNodeConfig(self.glb.logicalType, self.glb.connConfig)
+
+        # Get a chat to interface
+        self._chat = Chat(self.glb.logicalType, config, bag)
+
+    def endGlobal(self):
+        self._chat = None
+
+    def _format_error(self, status, etype, emsg, fallback: str) -> str:
+        """Compose a user-facing error string.
+
+        - If a numeric HTTP status/code is available, prefix it as "Error <status>:".
+        - Then include provider error type and message when present.
+        - If no structured fields are available, return the fallback message as-is.
+        - Whitespace is normalized to a single line; content is not truncated.
+        """
+        parts: list[str] = []
+        if status is not None:
+            parts.append(f'Error {status}:')
+        if etype:
+            parts.append(str(etype))
+        if emsg:
+            if etype:
+                parts.append('-')
+            parts.append(str(emsg))
+        message = ' '.join(parts) if parts else fallback
+        return re.sub(r'\s+', ' ', message).strip()

nodes/src/nodes/llm_nemotron/IInstance.py

@@ -0,0 +1,28 @@
+# =============================================================================
+# MIT License
+# Copyright (c) 2026 Aparavi Software AG
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+# =============================================================================
+
+from ai.common.llm_base import LLMBase
+
+
+class IInstance(LLMBase):
+    pass

nodes/src/nodes/llm_nemotron/README.md

@@ -0,0 +1,93 @@
+# NVIDIA Nemotron LLM Node (`llm_nemotron`)
+
+Connects RocketRide pipelines to NVIDIA's **Nemotron** family of open-weight
+reasoning models, served through NVIDIA's OpenAI-compatible cloud API
+([build.nvidia.com](https://build.nvidia.com)) or self-hosted with NIM
+containers, vLLM, or SGLang.
+
+- **Lane:** `questions → answers`
+- **Endpoint (cloud):** `https://integrate.api.nvidia.com/v1`
+- **Endpoint (local, default):** `http://localhost:8000/v1`
+
+## Models
+
+The Nemotron 3 generation are hybrid Mamba-Transformer MoE reasoning models
+with configurable thinking budgets.
+
+| Profile | Model ID | Context | Notes |
+| --- | --- | --- | --- |
+| Nemotron 3 Super 120B *(default)* | `nvidia/nemotron-3-super-120b-a12b` | 1M | Best efficiency/accuracy balance |
+| Nemotron 3 Ultra 550B | `nvidia/nemotron-3-ultra-550b-a55b` | 1M | Frontier reasoning, 55B active params |
+| Nemotron 3 Nano 30B | `nvidia/nemotron-3-nano-30b-a3b` | 256K | Fast sub-agent tier, 3B active params |
+| Custom Model | (user-defined) | (user-defined) | Any model on an OpenAI-compatible endpoint |
+
+The multimodal variants (Nano Omni, Nano VL) and the Retriever / Parse /
+Speech / Safety lines are out of scope for this chat node.
+
+## Reasoning output
+
+Nemotron 3 models are reasoning models. When the OpenAI-compatible endpoint
+returns chain-of-thought wrapped in `<think>...</think>` inside `content`, the
+node strips it so downstream nodes only see the final answer (the
+`llm_minimax` pattern). Budget generous output tokens for reasoning-heavy
+prompts.
+
+## Local deploy
+
+Self-hosted **NIM containers serve the same model IDs as the hosted API**, so
+the "(Local)" profiles reuse the cloud model strings against
+`http://localhost:8000/v1`:
+
+```sh
+docker run --gpus all -p 8000:8000 nvcr.io/nim/nvidia/nemotron-3-nano-30b-a3b:latest
+```
+
+vLLM / SGLang users can serve the open weights from HuggingFace instead; set
+`--served-model-name` to the profile's model ID (or use a Custom profile with
+your own name). Nemotron 3 weights, training data, and recipes are openly
+published; Nano runs on a single high-memory GPU, Super and Ultra need
+multi-GPU servers.
+
+## Authentication
+
+Cloud profiles require an NVIDIA API key (`nvapi-...`, from
+[build.nvidia.com](https://build.nvidia.com)) in `apikey`. The key requirement
+is enforced by base-URL match: if `serverbase` contains `api.nvidia.com` and
+no key is set, the node raises `NVIDIA API key is required for cloud
+profiles.` at startup. Key format is not validated beyond presence.
+
+Local profiles have no `apikey` field; local OpenAI-compatible servers accept
+any token, so the node passes a dummy key (`sk-local-dummy-key`).
+
+## Model sync
+
+Profiles are maintained by the `sync_models` tooling (`llm_nemotron`
+provider, `ROCKETRIDE_NVIDIA_KEY`). NVIDIA's `/v1/models` endpoint lists the
+full multi-vendor build.nvidia.com catalog (Llama, GLM, Kimi, ...), so the
+sync config filters to the `nvidia/*nemotron*` chat models only.
+
+---
+
+<!-- ROCKETRIDE:GENERATED:PARAMS START -->
+<!-- Generated by nodes:docs-generate. Do not edit by hand. -->
+
+## Schema
+
+| Field | Type | Description | Default |
+|---|---|---|---|
+| `model` | `string` | **Model**<br/>NVIDIA Nemotron model |  |
+| `modelTotalTokens` | `number` | **Tokens**<br/>Total Tokens |  |
+| `nemotron.profile` | `string` | **Model**<br/>NVIDIA Nemotron LLM model | `"nemotron-3-super"` |
+| `nemotron.serverbase` | `string` | **Server base URL**<br/>OpenAI-compatible base URL for the Nemotron endpoint (e.g. https://integrate.api.nvidia.com/v1 for NVIDIA's cloud API, http://localhost:8000/v1 for a self-hosted NIM / vLLM server). | `"https://integrate.api.nvidia.com/v1"` |
+
+## Dependencies
+
+- `openai`
+- `langchain-openai`
+- `langchain-core`
+- `langchain`
+
+## Source
+
+[<svg viewBox="0 0 16 16" width="15" height="15" fill="currentColor" aria-hidden="true" style="vertical-align:-0.15em;margin-right:0.35em"><path d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"/></svg> View source](https://github.com/rocketride-org/rocketride-server/tree/develop/nodes/src/nodes/llm_nemotron)
+<!-- ROCKETRIDE:GENERATED:PARAMS END -->

nodes/src/nodes/llm_nemotron/__init__.py

@@ -0,0 +1,41 @@
+# =============================================================================
+# MIT License
+# Copyright (c) 2026 Aparavi Software AG
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+# =============================================================================
+
+from .IGlobal import IGlobal
+from .IInstance import IInstance
+
+
+def getChat():
+    """
+    Get the Chat class from the module.
+    """
+    from .nemotron import Chat
+
+    return Chat
+
+
+__all__ = [
+    'IGlobal',
+    'IInstance',
+    'getChat',
+]