drop

Commit files to a GitHub repo from a static page. No server, no backend.

How it works

GitHub Pages is static hosting, so there is nothing to POST to. Instead the browser talks to the GitHub Contents API directly:

                       blob  blob  blob        one POST each
                          \   |   /
                           v  v  v
[browser] ---------------> tree ---> commit ---> PATCH ref ---> [main]
                                                                   |
[sidebar + table] <--GET /git/trees/main?recursive=1--------------- |
[preview]         <--raw.githubusercontent.com/.../{key}?v={sha}----+

Uploading writes git's plumbing directly: a blob per file, one tree, one commit, then move the ref. Nothing is visible until that last PATCH, so a queue of files either all lands or none of it does.

Three decisions worth knowing about:

• No index.json. The Git Trees API returns the entire bucket — every blob with its sha and size — in one request. Prefix navigation is then pure local filtering, so clicking into a folder costs zero requests. There is no index to drift out of sync with reality.
• The bucket root lives inside public/. That is the only directory output: export copies into the build, and being in the build is what gets a file served by Pages with a real Content-Type. An .html under public/ opens as a page; the same file served from raw is text/plain forever, because a domain that let anyone upload executable HTML would be a hole in GitHub. Move the bucket root outside public/ and objects still commit — they just stop being publishable, and the panel says so.
• Two URLs, two jobs. Previews read raw.githubusercontent.com because it is live the instant the commit lands, with the blob SHA as ?v= to defeat the ~5 minute CDN TTL. The shareable link is the Pages URL, which appears ~1–2 minutes later once Actions redeploys.
• Uploads trigger a rebuild. That is the point — Pages has to republish for the file to be reachable. Actions is free on public repos, so this costs time, not money.
• Folders are .gitkeep carriers. git can't represent an empty directory, so + folder commits a hidden .gitkeep. The S3 console does the same thing with a zero-byte key. Listings filter it out.

Constraints

Max file size	25 MB. The API allows 100 MB, but base64-in-JSON holds the file in browser memory twice
Requests per upload	N blobs + 4. Blobs go up in parallel; the ref moves once
Repo visibility	Public. Private repos need a token on the raw URL, which the <img> and <iframe> tags can't send
Time to a public link	~1–2 min after commit, once the Actions deploy finishes
Rate limit	5,000 requests/hour with a token
Who can upload	Only someone holding a write token — i.e. you