Cloud Storage Connectors in AACFlow: S3, Azure Blob, and AI Document Pipelines

Dropbox and Box — Common in creative and legal teams. AACFlow can download files, upload processed output, and move files between folders.

Yandex 360 Disk and Yandex Object Storage — The Russian-market equivalents, covered in detail below.

Selectel and VK Cloud — S3-compatible Russian cloud providers used for FZ-152 data localization.

SFTP — The universal fallback for enterprise file shares, legacy FTP servers, and on-premise storage that predates cloud. AACFlow's SFTP connector uploads and downloads over the SSH File Transfer Protocol, bridging old infrastructure to modern AI workflows.

Building an AI Document Processing Pipeline with S3

The most common pattern in AACFlow's storage connectors is a trigger-driven pipeline: a new file appears in a bucket, and that event starts an AI workflow automatically.

Invoice processing pipeline:

An accounts-payable team uploads PDF invoices to S3. Within seconds, AACFlow has extracted every line item, normalized the schema, written the data to Google Sheets, and archived the original file. No one touches the data between upload and the spreadsheet update.

Image analysis pipeline:

An e-commerce team uploads product photos. AACFlow generates descriptions and tags for search indexing, stores the metadata in PostgreSQL, and produces a thumbnail — automatically, for every image.

Report delivery pipeline:

The S3 Generate Presigned URL operation is what makes this elegant: the file stays private in S3, but the client gets a time-limited download link without needing AWS credentials. AACFlow handles the URL generation as a first-class workflow step.

How AACFlow Handles Files Inside Workflows

AACFlow treats files as typed workflow values that flow between blocks. The file-upload subBlock lets users attach files directly in the workflow canvas — PDFs, images, spreadsheets — and those files travel as structured references through the DAG.

When an S3 block downloads a file, it produces a UserFile object containing the storage reference and MIME type. The next block in the workflow receives that object and can pass it directly to an AI processing block — Textract for document extraction, Reducto for PDF parsing, or a vision model for image analysis — without needing to re-specify the file source.

For AI-generated file output, the FileToolProcessor handles storing the result back to cloud storage. An AI Agent block that generates a PPTX produces file bytes; the FileToolProcessor uploads those bytes to S3 (or any other configured storage target) and returns the storage reference for downstream steps to use.

File Output: Storing AI-Generated Documents

AI workflows often generate documents that need to live somewhere persistent. AACFlow's storage connectors handle the output side of the pipeline just as cleanly as the input side.

Generated PDF reports — An AI agent drafts a financial report. AACFlow uploads the PDF to S3, generates a presigned URL, and sends it by email. The file is in S3 for archival; the client gets a clean download link.

Extracted structured data — Reducto parses a complex document. The structured JSON output uploads to S3 alongside the original, creating an audit trail of what was extracted from what.

Processed media — A transcription workflow processes audio files. The transcript text file uploads to the same S3 folder as the source audio, paired by filename convention.

The Copy Object and Move Object operations in the S3 connector handle the archiving step cleanly: after processing, the source file moves from an inbox prefix to an archive prefix, preventing the trigger from firing again on the same file.

Russian Market: Data Sovereignty with Yandex Object Storage

For organizations operating under Russian law, FZ-152 (the personal data localization law) requires that personal data of Russian citizens be stored on servers physically located in Russia. Generic AWS or Azure storage in Frankfurt or Dublin does not satisfy this requirement.

AACFlow's Yandex Object Storage connector solves this directly. Yandex Object Storage implements the S3-compatible API — the same connector interface, the same operations (Upload, Download, List, Delete, Presigned URL, Copy), but with storage on Yandex's Russian infrastructure. Teams that already know how to use the S3 connector in AACFlow can switch to Yandex Object Storage by changing a single credential — the workflow logic stays identical.

Selectel and VK Cloud provide the same S3-compatible API and Russian data residency. AACFlow supports both as storage targets using the same connector pattern.

Compliance-driven pipeline example:

The entire pipeline keeps personal data within Russian jurisdiction. The audit trail is in a Russian PostgreSQL instance. The document is in Yandex Object Storage. Nothing leaves Russia.

Conclusion

Cloud storage is where business data lives before AI can act on it. AACFlow's storage connectors — AWS S3, Azure Blob, Google Drive, OneDrive, Dropbox, Box, Yandex Object Storage, Selectel, VK Cloud, SFTP — turn every bucket and folder into a potential AI workflow trigger or output target.

The patterns here scale from a single-team invoice processor to an enterprise document pipeline handling thousands of files per hour. The S3 trigger starts the workflow. The AI block does the work. The storage connector puts the result where it belongs.

Pick the bucket that holds your most time-consuming document type. Add a trigger. Add an AI processing step. The pipeline runs itself from there.