# FitLLM (fitllm.run)

Estimates whether a chosen model, quantisation and context length fit in a GPU's VRAM or a Mac's unified memory.

- Atlas card: https://inite.ai/en/atlas/fitllm.run
- Site: https://fitllm.run
- For: people checking whether a local LLM fits their NVIDIA, AMD or Mac memory before downloading
- Categories: commerce
- Use 47/100 · alive 100/100

## Used for

- check whether a model fits in GPU VRAM
- compare KV cache quantization and context length settings
- estimate memory left after loading a Hugging Face model config

## Price

Not read yet.

## For developers

- OpenAPI: https://fitllm.run/openapi.json (2 paths)
- MCP server: run.fitllm/fitllm (namespace; open on 2026-10-15, 3 tools)
- Read 2026-10-15

## MCP server, measured daily

Server: https://fitllm.run/api/mcp, checked every day since 2026-10-19 (2 checks).

- Answered: measuring, 2 daily checks so far
- Median first answer: 183 ms
- Sign-in: none
- Context its tool definitions take: about 910 tokens (3 tools)
- Tool definitions: A, 100/100
- Method: https://inite.ai/en/atlas/mcp/report

## Your data

- Certifications stated: none
- Read 2026-10-15

---

Source: INITE Atlas, https://inite.ai/en/atlas/fitllm.run. JSON: https://inite.ai/api/atlas/fitllm.run. MCP: https://inite.ai/api/mcp (atlas_lookup).