§02//CANIRUNAI
LIVE2025
CanIRunAIMatch open-source LLMs to your hardware.
If you've ever tried to run a local LLM, you know the question: "will this model even fit on my machine?" CanIRunAI answers that in under two seconds.
§01/the problem
Problem.
There are thousands of open-source LLMs, each with dozens of quantization variants, and working out which one fits on a specific GPU (or CPU + RAM combo) used to mean reading spec sheets, cross-referencing VRAM requirements, and losing an hour in a HuggingFace rabbit hole.
§02/the approach
Approach.
CanIRunAI indexes the open-source LLM ecosystem, tracks every quant variant from GGUF to AWQ, and scores each model against your specific hardware. Tell it your GPU (or let it auto-detect), and it returns a ranked list of models that will actually run — with honest speed estimates and quality tradeoffs.
§03/the ingestion pipeline
Ingestion pipeline.
The part I'm most proud of: the freshness. A Python scraper runs weekly via GitHub Actions, pulling new model releases from HuggingFace, parsing configs, extracting quant variants, and pushing updates to Supabase. Claude Haiku sits in the middle of the pipeline normalizing messy data (every model card is formatted differently). The result is an index that stays current with zero manual work.
§04/what's next
What's next.
v2 is in flight — expanded model coverage, more granular hardware profiles (Apple Silicon unified memory is its own can of worms), and a side-by-side compare view so you can evaluate two models before committing to a download.
MODELS TRACKED BY FAMILY
Approximate distribution across the open-source ecosystem.
total: 1,284 models
§stack
where it's at
Live, used, and growing. The automated ingestion means the index keeps improving whether I'm writing code that week or not — which is exactly the property I want out of this kind of tool.