Start with a recipe.
Choose a bundled profile. Sparkbot prepares its runtime and verifies the model files.
BUILT FOR NVIDIA DGX SPARK Private preview
The home for
your DGX Spark.
Your models. Your hardware. Your local AI.
Bring it all together in one simple dashboard.
One command to get started. Your Spark does the rest.
curl -fsSL …/bootstrap.sh | bashPublic installation opens when the GitHub repository is made public.
Your SparksConnected. Under control.
Your modelsChoose. Install. Run.
SparkBotLLMLocal help, on your hardware.
Less setup.More possibilities.
One place for your local AI.
NVIDIA DGX SparkQwenGLMTensorFoldLiquidAI
A CLEARER VIEWEverything clicks
Everything clicks
when it’s in one place.
Spend less time switching between terminals and tools. Get a clear view of your setup, then take the next step with confidence.
Your Spark · port 10005
Actual dashboard
Know how your model is running.Aggregate decode, prefill, cache reuse and hardware — clearly presented.
See the whole picture.
Watch tokens, throughput and unified memory. Benchmark decode and prefill in a few clicks.
Get help that’s local.
Ask SparkBotLLM to inspect your setup. Review changes before they happen.
FIND YOUR NEXT MODELA good starting point.
A good starting point.
A lot of room to explore.
Five bundled launch profiles, plus Mia’s AI Lab’s collection of DGX Spark recipes. Explore what fits your hardware.
5 bundled profiles
MMeet Mia’s AI Lab
More recipes. Proper credit.
Browse 33 DGX Spark sources from Mia’s AI Lab. Recipes awaiting adaptation are labelled for review, so you always know what’s ready to run.
SparkbotPowered by SparkBotLLM
LocalRECORDED EXAMPLE
What MHz are our Sparks running at?
Here’s what your Sparks reported.
HEAD SPARK1,989 MHz
WORKER SPARK1,989 MHz
A startup service requests a 2,000 MHz cap. Measured clocks can sit just below the ceiling.
Read-only inspection · no settings changed
Your hardware.
Your conversation.
Your conversation.
MEET SPARKBOTLLMA little help.
A little help.
Right where you need it.
Your local assistant understands the job: help you get more from your Sparks, one clear step at a time.
- Inspect Docker, GPU clocks, memory and connections.
- Troubleshoot your setup with terminal tools.
- Review commands and administrator changes before they run.
- Keep model inference on your own hardware.
SparkBotLLM
Powered by LiquidAI LFM3 1.7B · QAD Q4
SMALL BY DEFAULTYour first download
Your first download
shouldn’t be a project.
The dashboard comes first. Add the local helper when you want it. Larger model runtimes download only when you select a recipe.
1
Start with the dashboard
FirstOne command. A local web app. No model weights.
2
Add SparkBotLLM
Optional9.2 MB native CPU runtime + 1.05 GB Q4 model.
3
Choose a model recipe
Your choiceSee the download and disk requirements before setup.
The native CPU helper needs no Docker or GPU. GPU model recipes use their own runtimes. Sizes are measured package sizes, not memory use.
YOUR SETUP, YOUR WAYSimple when you start.
Simple when you start.
Capable as you grow.
DECODE BENCHMARKmax 400 tokens
1248
Concurrent requests · structured, prose or code
Measure what matters.
Compare decode and prefill with clear tables. See aggregate throughput and estimate the API equivalent of your local usage.
YOUR EXPERIENCE
Plain-language steps. Reviewed commands. Help when you need it.
Room for experience.
Start with guidance, or use Expert mode for direct terminal commands and launch overrides. Assistant-proposed changes still need approval.
A better Sparkbot, when you’re ready.Check. Review. Update.
Stay up to date.
Check GitHub releases from Settings. Review an update before installing it. The dashboard restarts; model services stay running.
Public update checks become available when the repository is public.
MAKE YOUR SPARK FEEL AT HOMELess setup.
Less setup.
More local AI.
Open a terminal on your DGX Spark.
One command gets the dashboard running.
On your DGX Spark
$ curl -fsSL https://raw.githubusercontent.com/x00byte/DGX-Sparkbot/main/bootstrap.sh | bash01
Run the installer on your Spark.
02
Open the setup link and create your password.
03
Choose a recipe. Add local help when you’re ready.
Private preview: the public command activates when the GitHub repository is public. No domain purchase is needed to run the dashboard.
A FEW GOOD QUESTIONS
Before you begin.
Made to be approachable.
Clear about what it does.
What do I need to run Sparkbot?
A Linux arm64 DGX Spark and an internet connection for installation. The dashboard runs locally on port 10005. For two Sparks, you’ll also need the interconnect and an SSH connection between them. Setup checks help you find missing components.
Do I need Docker before installing?
No. The dashboard and native CPU helper work without Docker. Model recipes may need Docker and NVIDIA container support; Sparkbot can inspect those requirements and present an installation plan.
Is SparkBotLLM private?
Helper inference runs on your own Spark, on loopback port 10004. It uses LiquidAI’s LFM3 1.7B QAD Q4 checkpoint. Downloads and discovery contact their public sources; proposed terminal commands and configuration changes need your review.
Can I still use my terminal and existing models?
Yes. Sparkbot can monitor an existing OpenAI-compatible model service. Its launchers manage their own containers. Expert mode gives you direct terminal access and supported recipe overrides.
Are all community recipes one-click installs?
The five bundled profiles have custom Sparkbot launchers. Other entries in Mia’s AI Lab’s catalogue are source links for review. Hardware, patches and licenses differ, so they are not presented as verified deployments.
Is this an official NVIDIA product?
Sparkbot is an independent project maintained by x00byte for DGX Spark owners. NVIDIA DGX Spark is a supported hardware platform, not an endorsement.