lightcraft/crates/fetch
Claude cab6af8608
Fetch: say where the bearer token goes (#259)
The token_env comment said the token goes to the first mirror's host; it
goes to every configured mirror's own host (never to a redirect's).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01912WaK6Rmew3wxZuPyX7E8
2026-10-08 17:55:31 +00:00
..
src Fetch: say where the bearer token goes (#259) 2026-10-08 17:55:31 +00:00
Cargo.toml Fetch: the model downloader is its own crate, lightcraft-fetch 2026-10-07 12:07:24 -04:00
README.md Fetch: validate caller-supplied names, sizes and hashes; FileSpec borrows its strings; download_url for an exact address 2026-10-07 22:59:22 -04:00

README.md

lightcraft-fetch

Native model downloads in pure Rust (L0, Apache-2.0): HTTP/1.1 over std sockets and rustls with the RustCrypto provider, without ring, aws-lc, C or assembly build dependencies. The wasm32 crate is empty.

Call download with bounded file specifications and ordered mirrors, or download_url for an exact catalog URL (including its query). Each file has a local filename independent of the URL, a size cap and optional pinned size/SHA-256. Transfers resume pinned .part files, verify before renaming, report progress and check cancellation during reads. Run on a worker thread. No proxy support; DNS resolution uses the platform resolver. Tokens are off by default and never forwarded to a different redirect host.

Tests use local synthetic HTTP servers for redirects, resume, truncation, cancellation, stalls, hashes, size limits and hostile specifications; they make no internet requests.