Changelog

1.0.2 - 2026-04-10

Second patch release for weightless.

Fixes:

Moved the initial provider scan into the TUI so startup and refresh use the same in-app progress flow
Turned disk-scan into a true on-demand home-tree scan instead of a narrow project-folder sweep
Kept disk-scan status and layout stable while drilling in and back out
Stopped double-counting shared-cache and explicit-provider models by assigning each artifact to a single most-specific provider
Made disk-scan report only leftover models instead of re-reporting storage already claimed by explicit providers

First patch release for weightless.

Fixes:

Made scans cwd-independent so running weightless from ~ or a repo returns the same results
Added startup progress reporting before the TUI opens so long scans no longer look frozen
Changed the broad fallback scan into a lazy disk-scan provider that only runs when drilled into from Summary
Tightened ambiguous file detection so generic .bin files and unrelated assets under models/ are no longer treated as model weights
Added regression coverage for ambiguous file detection and cross-provider scan behavior

Initial stable release of weightless.

Highlights:

Added a terminal UI with Summary and Models views for browsing local model storage
Added JSON output for exact machine-readable scans and scripting
Added provider drill-down, refresh, created dates, and open/reveal actions
Grouped raw files into logical model rows instead of file-per-file output

Provider coverage:

Added support for major local model stores including Ollama, LM Studio, AnythingLLM, Draw Things, Upscayl, GPT4All, Jan, vLLM, Chrome Built-in AI, NVIDIA, and shared Hugging Face caches
Added attribution for llama.cpp models that use the shared Hugging Face cache
Added fallback discovery for common models, weights, and checkpoints directories

Distribution: