AiMatch

RigSpark

RigSpark checks local hardware, recommends AI models that fit, and runs them on the user’s computer. Developers and local-AI users can use it for chat, generation, and hardware compatibility checks across several runtimes.

Visit RigSpark Compare alternatives Own this tool? Claim it
RigSpark screenshot

Who it's for

  • Developers who want to check whether a specific local model fits before adding it to a CI workflow.
  • Mac users with Apple Silicon who want hardware-aware model recommendations and local chat or generation.
  • Local-AI users working with Ollama, llama.cpp, MLX, BitNet, or LM Studio who want one tool to rank and run models across those runtimes.

Not the right fit if…

  • Teams looking for cloud-hosted image or video generation: RigSpark’s image and video workflows run through the user’s own ComfyUI.

How it fits your workflow

Run `rigspark recommend` to have RigSpark assess local hardware and rank models with fit and speed verdicts. Use `rigspark can-run <model>` to check one model, including as a CI gate.

Choose a model and run `rigspark up <model>`. RigSpark downloads and verifies the weights, then serves the model on 127.0.0.1 for local chat.

For image and video workflows, connect the user’s own ComfyUI. The desktop app also includes chat, while voice features use local speech models or a local speech server.

Pricing

Free: The Community Edition is free and open source.

Community Edition

Free

  • · Open source under the MIT license
  • · Available for macOS and Windows
  • · Desktop app bundles llama.cpp

Prices checked on Nov 11, 2026 from the vendor's site. They can change; confirm before you buy.

Key features

Hardware-aware model recommendations
The `rigspark recommend` command ranks catalog models against the computer’s hardware and gives each a yes, slow, or no verdict, with estimated tokens per second when available and a reason when a model will not fit.
Model download and integrity checks
RigSpark downloads selected model weights, checks them against catalog SHA-256 digests, and refuses weights that do not match.
Multiple local runtimes
The listed runtimes are Ollama, llama.cpp, MLX, BitNet, and LM Studio. The page describes LM Studio as attach-only and BitNet as CPU-based.
Local text chat
Users can chat, write, and work with code using a model selected for their hardware. RigSpark can start a model and serve it on 127.0.0.1.
Image and video generation
RigSpark can run image and short-video workflows through the user’s own ComfyUI, using pinned and SHA-256-verified model weights and built-in workflows.
Voice features
The page lists spoken replies using Kokoro or Chatterbox and dictation using Whisper. Spoken replies require Apple silicon; PCs attach to a local speech server.

Works with

  • Ollama
  • llama.cpp
  • MLX
  • BitNet
  • LM Studio
  • ComfyUI
  • MCP servers
  • Hugging Face

Limitations to know

  • Image and video generation run through the user’s own ComfyUI installation.
  • Spoken replies require Apple silicon; PCs need to attach to a local speech server.
  • The page says CUDA is unverified for Ming Design INT8.
  • Text in generated images may be garbled.
  • The container is an older CLI image and sees only the container’s resources.
  • The page says Windows SmartScreen may warn about the installer, and signed Windows builds are still being set up.

Getting started

rigspark recommend

The Community Edition desktop supports macOS 12 or later on Apple Silicon; downloads are also listed for macOS Intel and Windows x64. Command-line options are listed for macOS, Linux, and Windows.

  1. Download the Community Edition desktop app for macOS or Windows, or select a listed command-line installation method.
  2. Run `rigspark recommend` to view model rankings for the computer.
  3. Run `rigspark up <model>` to download, verify, and start a selected model.
  4. Run `rigspark chat` to chat with the running model.

Frequently asked questions

How do I find out which LLMs my computer can run?

Install RigSpark and run `rigspark recommend`. It detects GPU, VRAM, RAM, and free disk, ranks catalog models with yes, slow, or no verdicts, and gives reasons for models that will not fit. Use `rigspark can-run <model>` to check one model.

Can I run local LLMs on a Mac with Apple Silicon?

Yes. RigSpark sizes models against unified system memory and, for models with an MLX build, automatically selects MLX before Ollama and then llama.cpp.

Can RigSpark generate images and videos locally?

Yes. With the user’s own ComfyUI running on 127.0.0.1, RigSpark can install pinned, SHA-256-verified FLUX.1 schnell or Wan 2.1 weights and run a built-in workflow. The GUI also offers Image and Video composer modes.

Does RigSpark send my data anywhere?

The page says recommendations, `can-run`, and `doctor` make no network calls and use a verified local catalog. An explicit catalog update downloads a signed snapshot; normal startup does not. Model servers and the browser workspace bind to 127.0.0.1.

Is RigSpark the same project as llmup?

Yes. The page says RigSpark 2.0 renamed llmup, and the `llmup` command remains as a compatibility alias.