NemoClaw on Jetson Orin Super with jetpack 7.2

Robotics & Edge Computing Jetson Systems Jetson Orin Nano
,
1

Hello,

I updated firmware and successfully flashed and booted my Jetson Orin Nano on jetpack 7.2. Has anyone been able to get a simple local NemoClaw agent running on this system. I tried the following simple instructions:
curl -fsSL https://ollama.com/install.sh | sh (selected Ollama qwen2.5:3b)
nemoclaw mysandbox connect
openclaw tui

A dialog box opens in the terminal but it times out or crashes with the simplest querry. Is there a guide I can refer to on how to demo a simple local agent on the Jetson Orin Nano on jetpack 7.2

2

*** Please note that this reply is generated by LLM automatically ***

Unfortunately, I couldn’t find any specific information about running a NemoClaw agent on the Jetson Orin Nano in the provided context. However, I can try to provide some general guidance and potential troubleshooting steps.

Firstly, it’s great that you were able to update the firmware and successfully flash and boot your Jetson Orin Nano on JetPack 7.2.

To run a NemoClaw agent on the Jetson Orin Nano, you may want to try the following steps:

  1. Verify the installation: Double-check that the installation script completed successfully and that the NemoClaw agent is installed correctly. You can try running the installation script again or checking the installation logs for any errors.
  2. Check the system requirements: Ensure that your Jetson Orin Nano meets the system requirements for running the NemoClaw agent. This may include checking the version of JetPack, the Linux distribution, and any other dependencies required by the agent.
  3. Configure the agent: After installation, you may need to configure the NemoClaw agent to connect to your sandbox environment. Check the documentation for the agent to see if there are any specific configuration steps required.
  4. Troubleshoot the connection: If the agent is timing out or crashing when trying to connect to the sandbox environment, try checking the network connection and firewall settings on your Jetson Orin Nano. Ensure that the necessary ports are open and that the network connection is stable.
  5. Check for compatibility issues: If you’re still experiencing issues, it’s possible that there may be compatibility issues between the NemoClaw agent and the Jetson Orin Nano. You may want to try checking the agent’s documentation or reaching out to the developer community for support.

In terms of guides or documentation, I couldn’t find any specific resources that provide step-by-step instructions for running a NemoClaw agent on the Jetson Orin Nano. However, you may want to try checking the following resources:

  • The NemoClaw agent documentation: Check the official documentation for the NemoClaw agent to see if there are any specific instructions or guides for running the agent on the Jetson Orin Nano.
  • The Jetson Orin Nano documentation: Check the official documentation for the Jetson Orin Nano to see if there are any specific guides or instructions for running AI or machine learning workloads on the device.
  • Online forums and communities: Try searching online forums and communities, such as the NVIDIA developer forums or Reddit, to see if other users have experienced similar issues or have any advice or guidance to offer.

I hope these suggestions are helpful. If you have any further questions or need more specific guidance, please let me know.

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***

4

Please refer to NemoClaw on Jetson | Jetson AI Lab

6

Hi,

Nemoclaw can be installed via the command below directly:

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

The script also installs ollama (if ollama is selected) so you don’t need to manually install it first.
Thanks.

7

Thank you for your response. Yes, I followed the instructions here precisely with a freshly flashed Jetpack 7.2 on a Jetson Orin Nano, and with the jetpack installed:

The specific steps I used were:
sudo apt update && sudo apt install -y curl
curl -fsSL https://ollama.com/install.sh | sh
ollama pull nemotron-3-nano:4b
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
[In dialog, I chose Ollama as the model, and the model pulled above]
[I did not choose a messaging platform, intending to use terminal for simplicity]
source /home/jetson/.bashrc
nemoclaw my-assistant connect
openclaw tui

RESULT: terminal dialog opened with warning. Autocompaction could not recover this turn. Please use /compact or use /new. Increasy your compactioon buffer by setting agents.defaults.compaction.reserveTokensFloor to 20,000 or higher.

The dialog box would let me type a prompt but would never respond.
For your information, I could run:
ollama run nemotron-3-nano:4b
And dialog box opens, and I can give prompts and receive responses. So at this point:
I can run Ollama directly
I can not get it to work under nemoclaw.

I would appreciate direction.
Thanks!
Paul McWhorter

8

Hi,

We are checking this issue and will get back to you soon.
Could you share the Docker logs of your Nomoclaw container as well?

$ docker ps -a # find container ID
$ docker logs [ID]

Thanks.

9

Thank you for your response. The Orin I was working on becamed bricked in the process, so no access to the log files. I do hope you can give guidance on getting nemoclaw working on jetpack 7.2 on Jetson Orin Nano. I appreciate your help.
Paul McWhorter

10

Hi,

Thanks for the feedback.
We are double-checking this issue. Will get back to you later.

Thanks.

11

Hi,

We can reproduce the issue with Nemoclaw+Ollama on Orin+JetPack 7.2.
Will get back to you soon.

Thanks.

12

Any updates on this? I am facing the same issue.

14

Hi, all

There are some issues from Ollama:

  1. Skip GPU discovery → fallback to CPU
  2. Sandbox was configured with contextWindow=4096 → caused “Context overflow” error

After applying the settings below, we can get NemoClaw running without issue.

  1. Edit /etc/systemd/system/ollama.service.d/override.conf
[Service]
Environment="OLLAMA_HOST=127.0.0.1:11434"
Environment="OLLAMA_CONTEXT_LENGTH=32768"
Environment="OLLAMA_IGPU_ENABLE=1"
Environment="GGML_BACKEND_PATH=/usr/local/lib/ollama/cuda_v13/libggml-cuda.so"
Environment="LD_LIBRARY_PATH=/usr/local/lib/ollama:/usr/local/lib/ollama/cuda_v13"

$ sudo systemctl daemon-reload
$ sudo systemctl restart ollama
  1. Edit openclaw setting
$ nemoclaw my-assistant connect
# sed -i.bak 's/"contextWindow"[[:space:]]*:[[:space:]]*4096/"contextWindow": 32768/' .openclaw/openclaw.json
# sha256sum .openclaw/openclaw.json > .openclaw/.config-hash 
# openclaw tui

Thanks

15

Thank you so much for your solution, I will proceed with this method.

Paul McWhorter

16

I can confirm the following works, and results in ollama running on the GPU under jetpack 7.2 on the Jetson Orin Nano.
Step 1: Follow the install instructions at the top of this thread.
Step 2: Then do the AastaLLL edits specified above.

17

Hi,

Thanks for the update.
Good to know it works now.