Most AI Agents and LLMs expect clean data via modern APIs (JSON, REST). But your critical operational data (manufacturing logs, mainframe exports, PLC outputs, and surveillance footage) is locked inside legacy hardware that only speaks FTP or SFTP.
We hit this wall while building RAG pipelines for industrial sectors: The AI connects easily to Notion, but it cannot talk to a 1990s PLC or a security camera system.
Traditional managed file transfer (MFT) solutions focus on human-to-human workflows: shared folders, email notifications, and manual downloads. AI Agents and RAG pipelines need event-driven triggers and direct storage access, not a folder to watch.
The Old Way: Polling and Scripts
We kept seeing DevOps engineers patch this gap with "glue code":
- Writing a Python script to poll the FTP server every hour.
- Downloading the file locally.
- Uploading it to S3.
This is slow and expensive. You are paying for compute time just to check for empty directories, and your AI is always minutes or hours behind reality.
The Solution: The "Event-Driven" Bridge
Rilavek acts as the "Universal Adapter" for this legacy-to-modern gap.
- Step 1 (Ingest): Your legacy hardware uploads via standard SFTP (Port 2222) or FTPS. It thinks it is talking to a dumb file server.
- Step 2 (Transform): Rilavek streams the data directly to your S3 bucket (or MinIO/Weaviate backend).
- Step 3 (Trigger): The moment the file lands, Rilavek fires a JSON payload to your AI Agent (e.g., LangChain or AutoGPT) via Webhooks.
Benefit: No polling. Your pipeline processes new data within seconds of upload, not hours.
Security for Sensitive Datasets
AI data is often proprietary. Unlike "Hosted MFTs" that store the data at rest, Rilavek uses a Zero-Knowledge Pass-Through architecture.
We stream your data in-memory directly to your storage. Your proprietary training data never sits on our disk, ensuring compliance with strict data governance policies.
Tutorial: Build a Live RAG Pipeline in 5 Minutes
Here is how to set up a pipeline that embeds PDF reports from a mainframe instantly:
1. Create a Pipe
Point Rilavek to your S3 bucket where your Vector Database pulls source documents.
2. Add a Webhook URL
In your Rilavek Pipe settings, add the endpoint for your embedding service (e.g., a FastAPI endpoint or a Lambda function).
For the full payload schema, check out our Webhook Documentation.
3. Upload a Test File
Connect your legacy client (or curl for testing) and upload a file.
curl -T report.pdf -u "MySender@pipe_abc:my-password" ftp://ftp.rilavek.com --ssl
4. Watch the Trigger
Your Webhook endpoint receives the event instantly. Your AI agent can now fetch s3://my-rag-bucket/input/report.pdf and start the embedding process immediately.
Why This Matters
Stop building fragile polling scripts. Rilavek bridges your SFTP/FTP sources directly into your event-driven AI architecture. No middleware, no cron jobs. Legacy files become real-time data streams for your RAG pipeline.