What causes "Binary content not supported" in langflow-ai/langflow?

Use a binary-capable path (the download/attachment flow or the storage service) for non-text artifacts.

langflow-ai/langflow · warning · HTTPException

Binary content not supported

Error message

Binary content not supported

What it means

The agentic file router serves text content only: after the binary sniff, _looks_binary(head) checks for a null byte in the first BINARY_SNIFF_BYTES. If found, it raises HTTP 415 'Binary content not supported' — the same heuristic the FS tool uses — because returning raw bytes through this text-oriented endpoint would corrupt responses.

Source

Thrown at src/backend/base/langflow/agentic/api/files_router.py:175

            resolved.exists(),
            resolved.is_dir(),
        )
        raise HTTPException(status_code=404, detail="Not found")

    try:
        size = resolved.stat().st_size
    except OSError:
        raise HTTPException(status_code=404, detail="Not found") from None
    if size > MAX_FILE_SIZE_BYTES:
        raise HTTPException(status_code=413, detail="File too large")

    # Binary sniff before the full read — same heuristic the FS tool uses.
    try:
        head = _read_head_no_follow(resolved, BINARY_SNIFF_BYTES)
    except (OSError, PermissionError):
        raise HTTPException(status_code=404, detail="Not found") from None
    if _looks_binary(head):
        raise HTTPException(status_code=415, detail="Binary content not supported")

    try:
        body_bytes = _read_bytes_no_follow(resolved)
    except (OSError, PermissionError):
        raise HTTPException(status_code=404, detail="Not found") from None

    # Audit log — relative path only, never the absolute sandbox-resolved path.
    logger.info(
        "agentic.files.read user_id=%s path=%s size=%d download=%s",
        current_user.id,
        path,
        len(body_bytes),
        download,
    )

    if download:
        filename = _safe_filename_for_disposition(path) or "file"
        return Response(

View on GitHub (pinned to 976ec789d2)

Solutions

Use a binary-capable path (the download/attachment flow or the storage service) for non-text artifacts.
Instruct the agent to write textual formats (markdown/json/csv) for files intended to be read via this endpoint.
Treat 415 as final for that file — retrying will not change the sniff result.

Defensive patterns

Strategy: validation

Validate before calling

SNIFF = 8192
with open(sandbox_path, "rb") as f:
    looks_binary = b"\x00" in f.read(SNIFF)
if looks_binary:
    use_download_flow(sandbox_path)

Try / catch

if resp.status_code == 415:
    switch_to_binary_transfer(path)  # sniff is deterministic; retries are useless

Prevention

Reserve this endpoint for text formats (md/json/csv/txt).
Instruct agents to persist artifacts as text when they will be inlined.
Use the attachment/download path for images, archives, and serialized objects.

When it happens

Trigger: GET /api/v1/agentic/files on a .png, .zip, .sqlite, .pkl, or any file with a NUL byte in its first 8 KiB.

Common situations: Agent saves plots/screenshots or serialized artifacts in the sandbox and the UI tries to inline-preview them; CSV with embedded NULs from a bad export.

Related errors

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/318eae074eb0e22f. Report an issue: GitHub.