DevelopersIntegrationsNVIDIA NemoClaw

NVIDIA NemoClaw with ARK

Keep NemoClaw's sandboxing and guardrails, and send inference to open models served in the EU or on your own ARK deployment.

By NVIDIAApache 2.0 licenceSourceVendor docs

NVIDIA NemoClaw: NVIDIA's open-source reference stack for running AI agents such as OpenClaw and Hermes more safely inside OpenShell sandboxes. NemoClaw is an alpha project.

Before you start

You need an ARK API key: sign in to ARK Cloud and copy it from the dashboard, or use a key from your own ARK deployment and its base URL. The examples use arklabsai/Qwen3.6-35B-A3B; for the hardest tasks use arklabsai/Qwen3.5-122B-A10B. Any chat model from the catalogue works.

1. Install

Install
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash

2. Connect it to ARK

Onboard NemoClaw against ARK as a custom OpenAI-compatible endpoint. Set the model limits first: NemoClaw bakes them in at build time.

Onboarding
export NEMOCLAW_CONTEXT_WINDOW=262144
export NEMOCLAW_MAX_TOKENS=8192
export COMPATIBLE_API_KEY="your-ark-api-key"

NEMOCLAW_PREFERRED_API=openai-completions \
  NEMOCLAW_PROVIDER=custom \
  NEMOCLAW_ENDPOINT_URL=https://api.ark-labs.cloud/api/v1 \
  NEMOCLAW_MODEL=arklabsai/Qwen3.6-35B-A3B \
  COMPATIBLE_API_KEY="$COMPATIBLE_API_KEY" \
  nemoclaw onboard --non-interactive

Interactively, run nemoclaw onboard, choose Other OpenAI-compatible endpoint and enter the base URL and model id. To switch an existing sandbox, use nemoclaw inference set --provider compatible-endpoint … --inference-api openai-completions.

Good to know

  • NEMOCLAW_PREFERRED_API=openai-completions skips NemoClaw's probe of /v1/responses, which ARK does not serve.
  • NemoClaw's defaults are 131,072 tokens of context and 4,096 output tokens; changing them later means recreating the sandbox.
  • NVIDIA lists custom OpenAI-compatible endpoints as tested with limitations; onboarding checks the API, tool calling and streaming.
  • These tools send the whole conversation on every turn. ARK's prompt cache speeds up the repeated part of each prompt; every prompt token is still billed.

Next

Read about tool calling and limits and capacity on ARK, or browse other integrations.

Configuration keys follow NVIDIA NemoClaw's official documentation as of October 2026. If the tool has changed since, its docs take precedence.