Launching today

canfit
The top command for people running local LLMs and apps
28 followers
The top command for people running local LLMs and apps
28 followers
canfit tells you if a local model or app can launch right now, and if not, what you can safely close to make room. System processes are never touched. Open-source Rust TUI/CLI for macOS, Windows and Linux, no telemetry.





Free
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt! 👋 I'm Federico, maker of canfit.
canfit was born out of frustration. While building my previous open-source project, Knemora, I was running local LLMs on my Mac alongside Docker, browsers and everything else, and memory kept running out. Every time I asked myself the same thing: what can I close safely, without killing something macOS needs?
So I built the tool I wanted:
Can I launch it?
Point canfit at a GGUF model or an app, and it tells you if it fits in the memory you have right now, including the GPU limit on Apple Silicon.
What should I close?
If it doesn't fit, canfit suggests the smallest set of apps to close. System processes are never suggested, and nothing is closed without your confirmation.
A friendlier top.
A live monitor that starts with a sentence ("Memory under pressure: Chrome is using 9 GB") and groups processes by app.
Watch. Launch a command through canfit and it measures its real usage, so the next estimate is better.
It's a single Rust binary for macOS, Windows and Linux, MIT licensed, with no telemetry.
Install it with:
This is 1.0.0, so I'd really love your feedback, especially if canfit says a model fits and it doesn't (or the other way round).
Tell me your setup and I'll make the estimates better.
Thanks for checking it out! 🙏
does it count docker desktop's vm as one blob or can it see whats inside? on my mac thats always the thing eating ram and closing chrome tabs barely helps lol
@bacelj One blob, honestly. canfit sees processes on the Mac, Linux, Windows, not inside Docker’s Linux VM, so it can’t tell you which container is the hungry one (docker stats is still the tool for that).
What it does with the blob:
Shows it as the culprit. On recent Docker Desktop the VM runs as com.apple.Virtualization.VirtualMachine, and the monitor lists it with its real size and marks it “caution: closing it stops it like pulling the plug”. So you see straight away that it’s the VM, not Chrome.
Reads what it’s allowed to take. canfit "Docker Desktop" reads the VM memory limit from Docker’s settings, so you know how much it can grow to, not just what it uses now.
Tells you how to stop it properly. In “what to close” it suggests docker desktop stop before any kill.
That’s also why closing tabs barely helps: the VM keeps can hold on to memory it has grabbed even after the containers go idle. Lowering the memory limit in Docker’s settings is often the real fix, and canfit "Docker Desktop" model.gguf tells you whether the two fit together.
Seeing per-container usage would mean talking to the Docker API.
If you like the project, a ⭐ on GitHub (github.com/federicolaterra/canfit) means a lot and really helps.
More features are coming in the next releases.
does the fit guess count the context growing as a chat gets longer, thats usually when mine falls over
@igorblink Yes, but it’s the worst case rather than a curve: canfit sizes the KV cache for the full context window, as if the chat were already at the limit, plus compute buffers that also scale with context. llama.cpp and Ollama allocate the whole cache up front, so the window you load with is what matters.
The catch is the default: without a flag canfit assumes 8,192 tokens (or the model’s own max if smaller).
If your chats run longer, tell it the real window:
The card always says which context it used.
To check against reality,
records the real peak of the whole process tree over the session and shows it next to the estimate.
Later,
lists it as “measured by watch”, and canfit my-chat answers from that measured peak instead of an estimate
(--no-save if you don’t want it kept).
If you like the project, a ⭐ on GitHub (github.com/federicolaterra/canfit) means a lot and really helps.
More features are coming in the next releases.