weightless

Find local AI model weights, VM/runtime stores, and LLM session files across desktop apps, shared caches, and project folders.

weightless is for the messy real world where Ollama, LM Studio, Hugging Face, Draw Things, Docker, Podman, Lima, Apple simulators, Claude, Codex, Copilot, Antigravity, OpenCode, and one-off repos all store heavy local files in different places. It gives you one interactive terminal UI plus a JSON mode for scripting and debugging.

Highlights

• Scans provider-specific model stores, virtual machine/runtime stores, and LLM session stores by default, with an optional on-demand disk-scan for broader model folders
• Groups raw files into logical models so sharded packages show up as one row
• Shows size, provider, category, created date, and path in JSON
• Lets you drill from Summary into provider-specific artifacts
• Adds dedicated tabs for Models, Virtual Machines, and LLM Sessions
• Refreshes in place with r
• Emits machine-readable JSON
• Keeps provider detection easy to extend in internal/providers/registry.go

Install

Install script:

curl -fsSL https://raw.githubusercontent.com/needle-tools/weightless/main/install.sh | bash

Specific version:

curl -fsSL https://raw.githubusercontent.com/needle-tools/weightless/main/install.sh | bash -s -- -s 1.1.0

Or download a release archive directly from GitHub Releases.

Run

weightless
weightless --json
weightless --version

Common flags:

weightless --providers ollama,lm-studio,huggingface
weightless --providers docker,podman,lima,apple-simulators
weightless --providers claude,codex,cursor,opencode
weightless --roots ~/work/models,/Volumes/FastSSD/models
weightless --min-size-mb 8

TUI

Keys:

• ← and → switch tabs
• enter or space drills into a provider from Summary
• o opens or reveals the selected item
• r refreshes the scan
• esc goes back from a drilled view
• q quits

Providers

Model coverage includes:

• ollama
• lm-studio
• anythingllm
• draw-things
• upscayl
• huggingface
• unsloth-studio
• jan
• gpt4all
• vllm
• node-llama-cpp
• llama.cpp shared-cache attribution
• chrome-built-in-ai
• nvidia
• text-generation-webui
• comfy
• stable-diffusion-webui
• invokeai
• disk-scan (lazy, on demand from Summary)

Virtual machine and runtime coverage includes:

• docker
• podman
• lima
• apple-simulators
• apple-simulator-runtimes
• android-emulator
• claude-vm
• codex-vm
• utm
• vercel-sandbox

LLM session coverage includes:

• claude
• codex
• copilot
• antigravity
• opencode
• cursor
• windsurf
• cline
• roo-code
• kilo-code
• aider
• gemini-cli
• qwen-code

JSON Output

Example shape:

{
  "categories": [
    {
      "category": "models",
      "artifacts": 31,
      "size_bytes": 111883059200,
      "size_human": "104.2 GiB"
    }
  ],
  "summary": [
    {
      "provider": "ollama",
      "artifacts": 2,
      "complete_artifacts": 2,
      "incomplete_artifacts": 0,
      "size_bytes": 7630497504,
      "size_human": "7.1 GiB"
    }
  ],
  "artifacts": [
    {
      "category": "models",
      "name": "qwen3.5:9b",
      "model_name": "qwen3.5:9b",
      "status": "complete",
      "primary_provider": "ollama",
      "path": "/Users/you/.ollama/models/blobs/sha256-...",
      "timestamp": "2026-04-08T09:15:00+02:00",
      "file_count": 1,
      "all_paths": [
        "/Users/you/.ollama/models/blobs/sha256-..."
      ]
    }
  ],
  "total_artifacts": 31,
  "total_size_human": "104.2 GiB"
}

Development

Build locally:

PATH=/opt/homebrew/bin:$PATH go build -o weightless .

Run from source:

go run .

Build from source with the installer:

./install.sh --build-from-source

Release

This repo is set up to publish GitHub Releases directly.

One-time maintainer setup:

1. Create the GitHub repo needle-tools/weightless.
2. Push main.

Publish a release:

git push origin main
git tag v1.1.0
git push origin v1.1.0

That release flow will:

• run CI
• build macOS, Linux, and Windows binaries
• publish GitHub Release assets
• generate checksums

Changelog

See CHANGELOG.md.