Roboflow + MinIO

Run an RF-DETR model over every image and video in a MinIO bucket with Roboflow Batch Processing, and get the predictions back as JSONL or CSV.

MinIO is S3-compatible object storage you run on your own servers. Roboflow Batch Processing (opens in a new tab) runs a Workflow over the images and videos in a MinIO bucket on Roboflow's cloud and returns the predictions as one JSONL or CSV file per job, without importing anything into your workspace.

The inference CLI talks to MinIO through its S3 API: you set your server's endpoint and access key, and the CLI lists the files that match an s3:// path and hands Roboflow presigned URLs. We recommend an RF-DETR Workflow: it detects the 80 COCO classes out of the box, or your own after fine-tuning, and runs on CPU.

How to connect MinIO to Roboflow

6 steps
  1. Build an RF-DETR Workflow

    In Roboflow, create a Workflow with an RF-DETR object detection model, either pre-trained on the 80 COCO classes or fine-tuned on your own, and copy its Workflow ID.

  2. Install the inference CLI

    Run `pip install 'inference-cli[cloud-storage]'` and set `ROBOFLOW_API_KEY` to your Roboflow API key.

  3. Point the CLI at MinIO

    Set `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` to a MinIO access key that can read the bucket, `AWS_ENDPOINT_URL` to your MinIO server's address and `AWS_REGION` to its region.

  4. Stage the files in your bucket

    Run `inference rf-cloud data-staging create-batch-of-images --batch-id my-batch --data-source cloud-storage --bucket-path 's3://my-bucket/images/**/*.jpg'`. For video, use `create-batch-of-videos` with a path to your video files.

  5. Run the Workflow

    Run `inference rf-cloud batch-processing process-images-with-workflow --batch-id my-batch --workflow-id <your-workflow-id> --machine-type cpu --aggregation-format csv`. Use `process-videos-with-workflow` for a video batch.

  6. Download the results

    Check progress with `inference rf-cloud batch-processing show-job-details --job-id <job-id>`, then run `inference rf-cloud data-staging export-batch --batch-id <job-id>-export --target-dir ./results`.

What teams build with MinIO

  • Audit an archive of inspection photos

    Count defects, parts or people across every image already sitting in MinIO, in one job.

  • Find moments in stored video

    Run detection over recorded footage at a set frame rate and get a row for every frame with a match.

  • Tag files before labeling

    Get predictions for a whole bucket so your team knows which files are worth labeling first.

Questions

Yes. Roboflow's cloud downloads each file through a presigned URL that points at your server, so the endpoint has to be reachable from the public internet. The same applies to a Datasource.

Yes, if you are happy to import the images into Roboflow. Connect the bucket as a Datasource in your workspace settings (choose S3 and set your MinIO endpoint), then add a Workflow run set to "On Sync" so every sync runs your Workflow over the newly imported images. The CLI steps on this page import nothing.

Yes. Connect the Roboflow MCP server to Claude Code, Cursor or Codex and paste: "Run my RF-DETR Workflow over every JPG in my MinIO bucket at s3://my-bucket/images/ with Roboflow Batch Processing on CPU, and save the results as CSV in ./results." The agent stages the files with the CLI on your machine, then starts and tracks the job.

No. RF-DETR runs on CPU, which is the Batch Processing default and costs fewer credits than GPU. Choose GPU for Workflows that chain several models or use large ones such as SAM.

Staged files and job results expire after about seven days, so download them with `export-batch` once the job finishes.

Why are companies choosing Roboflow?

The end-to-end platform for vision AI: create datasets, train models and deploy to production.

  • Flexible deployment modes

    Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.

  • Integrated workflow builder

    Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.

  • Model monitoring and alerts

    Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.

View all
Get started

Build on the Platform

For developers, engineers, and technical founders who want to get hands on. Try the free tier; the docs are open.

Bring it into your operation

For heads of AI, operations leaders, and enterprise teams. Bring a known problem, or work with us to find the first one worth taking on.