memv-files

memv-files

Required reading

1. docs/memv/sdk/files.md — upload API (client.upload.batch.create).
2. docs/memv/sdk/error-handling.md — upload-failure modes.
3. docs/memv/sdk/videos.md — video → use that skill instead (same upload.batch.create path, video skill has format guidance).

SDK surface (verified against docs/memv/sdk/files.md)

• client.upload.batch.create(space_id, files=[...], ...) — upload PDF / doc / image / audio
• client.upload.batch.get_status(batch_id) — poll for processing done (sync wrapper) / TS: getStatus
• client.files.list(space_id) — enumerate files in space

Hard rules

• Push raw files. No OCR / parse / chunk before upload — mem[v] handles extraction.
• space_id mandatory.
• upload.batch.create async. Returns batch_id. Poll get_status(batch_id) until done before assuming memories searchable.
• Parallelize via asyncio.gather with backoff per sdk/error-handling.md.

Skeleton (Python)

with open(file_path, "rb") as f:
    response = client.upload.batch.create(
        space_id=space_id,
        files=[f],
        metadata={"source": "user_upload", "filename": file_path.name},
    )
batch_id = response.batch_id

# poll until processing completes
while True:
    status = client.upload.batch.get_status(batch_id=batch_id)
    if status.state in ("completed", "failed"):
        break

Choose right method

• Video → client.upload.batch.create (skill memv-video-ingest for format guidance)
• PDF, doc, image, audio → client.upload.batch.create (this skill)
• Plain text already in memory → client.memories.add(content=...) (skill memv-add-memory)

Don't

• No client.memories.add(content=open(pdf).read()) — stuffs raw bytes as text. Use upload.batch.create.
• No pre-OCR. mem[v] extracts on its side.
• No assuming write searchable immediately after create() returns. Poll get_status first.