gr.Files — upload a whole batch, in one preset line

$ import csv, gradio as gr

def merge(files):
    rows, header = 0, None
    for p in files or []:
        with open(p, newline="", encoding="utf-8") as f:
            r = csv.reader(f)
            header = header or next(r)
            rows += len(list(r))
    return {"files": len(files or []), "data_rows": rows, "columns": header}

with gr.Blocks(title="batch-merge") as demo:
    gr.Markdown("# Batch CSV merger")
    f = gr.Files(label="Drop your CSVs", file_types=[".csv"], elem_id="f")
    b = gr.Button("Merge")
    o = gr.JSON(label="merged")
    b.click(merge, f, o)

demo.launch()

* Running on local URL:  http://127.0.0.1:7860

Renders a Blocks page: a Markdown H1, then a gr.Files drop zone labeled 'Drop your CSVs', a Button 'Merge', and a JSON output labeled 'merged'. The drop zone's inner text before upload is exactly 'Drop File Here - or - Click to Upload'.
Uploaded north.csv (32 bytes) and south.csv (23 bytes) from a real Chromium (Playwright). The component rendered two rows: 'north .csv 32.0 B download-icon remove-x' and 'south .csv 23.0 B download-icon remove-x'.
Your function receives the list ['/tmp/gradio/<hash>/north.csv', '/tmp/gradio/<hash>/south.csv'] — both under the SAME hash directory. Clicking Merge returned: { "files" : 2 , "data_rows" : 4 , "columns" : [ "region" , "sales" ] }.
Tested with gradio 6.18.0 / Python 3.13 on 2026-10-09.

Both files land inside one hash directory (the per-session cache), not next to each other by chance. Basename collisions do not help you distinguish them — open by full path, never by name.

$ import gradio as gr

with gr.Blocks() as demo:
    f = gr.File(type="binary", label="Any file", elem_id="f")
    out = gr.Textbox(label="header check")
    f.upload(
        lambda data: f"bytes={len(data)} head={data[:8]!r}" if data else "no file",
        f, out,
    )

demo.launch()

* Running on local URL:  http://127.0.0.1:7860

Renders a Blocks page with one single-file drop zone ('Any file') wired on the file-specific .upload() listener to a 'header check' Textbox.
Uploaded an 11-byte fake GIF (GIF89a + 0xff + five zero bytes). The input element had accept=None, i.e. no filter was applied to the file picker at all. The Textbox printed: bytes=11 head=b'GIF89a\xff\x00'.
type='binary' handed the function real bytes (type bytes, len 11) with no disk round-trip in your code; type='filepath' would have handed the same file as a str path instead.
Tested with gradio 6.18.0 / Python 3.13 on 2026-10-09.

Use .upload() here, not .click() on a Button — with binary payloads you want the check to run the instant the file lands, not after a second user action.

$ import gradio as gr, os

def pick(folder):
    if not folder:
        return "no folder selected"
    names = sorted(os.path.basename(p) for p in folder)
    exts = sorted({os.path.splitext(n)[1] for n in names})
    return f"{len(names)} files: {names[:3]} ... extensions: {exts}"

with gr.Blocks() as demo:
    d = gr.File(file_count="directory", label="Pick a folder", elem_id="f")
    out = gr.Textbox(label="scan")
    gr.Button("Scan").click(pick, d, out)

demo.launch()

* Running on local URL:  http://127.0.0.1:7860

Renders the same layout as example 2 but the drop zone is folder mode. Playwright confirmed the underlying input carries BOTH webkitdirectory and directory attributes — that is what opens the native folder picker.
Pointed it at a test directory containing a.txt, b.csv, c.png and a nested sub/ directory holding d.txt. The UI showed four rows (c.png, b.csv, a.txt, d.txt). The Textbox printed: "4 files: ['a.txt', 'b.csv', 'c.png'] ... extensions: ['.csv', '.png', '.txt']".
The folder was walked RECURSIVELY, but every path came back flattened into one directory: the nested sub/d.txt arrived as /tmp/gradio/<hash>/d.txt with no 'sub/' component. Tested with gradio 6.18.0 / Python 3.13 on 2026-10-09.

Directory mode is recursive and lossy in one specific way: the tree structure is discarded. If a nested file collides on basename with a top-level one, you cannot tell them apart from the paths alone.