• ⚡ Out-of-the-Box: No External Dependencies Required & Support Vercel/Render One-Click Deployment
• ⭐ Multiple File Types: Support Text, Pdf, Docx, Excel, Image, Audio etc.
• 📦 Multiple Storage Options: Base64, Local, S3, Cloudflare R2, Min IO, Telegram CDN etc.
• 🔍 OCR Support: Extract Text from Image (Require Paddle OCR API)
• 🔊 Audio Support: Convert Audio to Text (Require Azure Speech to Text Service)

• Text
• Image (require vision models)
• Audio (require Azure Speech to Text Service)
• Docx (not support .doc)
• Pdf
• Pptx (not support .ppt)
• Xlsx (support .xls)

Image: programzmh/chatnio-blob-service

docker run -p 8000:8000 programzmh/chatnio-blob-service

# with environment variables
# docker run -p 8000:8000 -e AZURE_SPEECH_KEY="..." -e AZURE_SPEECH_REGION="..." programzmh/chatnio-blob-service


# if you are using `local` storage type, you need to mount volume (/static) to the host
# docker run -p 8000:8000 -v /path/to/static:/static programzmh/chatnio-blob-service

Deploy to Render.com

Select Web Service and Docker Image, then input the image programzmh/chatnio-blob-service and click Create Web Service.

⭐ Render.com Includes Free 750 Hours of Usage per Month

The service will be running on http://localhost:8000

git clone --branch=main https://github.com/Deeptrain-Community/chatnio-blob-service
cd chatnio-blob-service

pip install -r requirements.txt
uvicorn main:app

# enable hot reload
# uvicorn main:app --reload

POST /upload Upload a file

{
    "file": "file",
    "model": "gpt-4-turbo-preview" // optional (for ocr models detection)
}

Response

{
  "status": true,
  "type": "pdf",
  "content": "...",
  "error": ""
}

• PDF_MAX_IMAGES: Max Images Extracted from a PDF File
  • 0: Never Extract Images
  • -1: Extract All Images
  • other: Extract Top N Images
  • Tips: The extracted images will be treated as a normal image file and directly processed.
• MAX_FILE_SIZE: Max Uploaded File Size MiB (Default: No Limit)
  • Tips: Size limit is also depend on the server configuration (e.g. Nginx/Apache Config, Vercel Free Plan Limit 5MB Body Size)
• CORS_ALLOW_ORIGINS: CORS Allow Origins (Default: *)
  • e.g.: http://localhost:3000,https://example.com
• AZURE_SPEECH_KEY: Azure Speech to Text Service Key (Required for Audio Support)
• AZURE_SPEECH_REGION: Azure Speech to Text Service Region (Required for Audio Support)

1. ✨ No Storage (Default)

   • No Storage Required & No External Dependencies
   • Base64 Encoding/Decoding
   • Support Serverless Deployment Without Storage (e.g. Vercel)
   • No Direct URL Access (Base64 not support models like gpt-4-all)

2. 📁 Local Storage

   • Require Server Environment (e.g. VPS, Docker)
   • Support Direct URL Access
   • Payless Storage Cost
   • Config:
     • set env STORAGE_TYPE to local (e.g. STORAGE_TYPE=local)
     • set env LOCAL_STORAGE_DOMAIN to your deployment domain (e.g. LOCAL_STORAGE_DOMAIN=http://blob-service.onrender.com)
     • if you are using Docker, you need to mount volume /static to the host (e.g. -v /path/to/static:/static)

3. 🚀 AWS S3

   • Payment Storage Cost
   • Support Direct URL Access
   • China Mainland User Friendly
   • Config:
     • set env STORAGE_TYPE to s3 (e.g. STORAGE_TYPE=s3)
     • set env S3_ACCESS_KEY to your AWS Access Key ID
     • set env S3_SECRET_KEY to your AWS Secret Access Key
     • set env S3_BUCKET to your AWS S3 Bucket Name
     • set env S3_REGION to your AWS S3 Region

4. 🔔 Cloudflare R2

   • Free Storage Quota (10GB Storage & Zero Outbound Cost)
   • Support Direct URL Access
   • Config (S3 Compatible):
     • set env STORAGE_TYPE to s3 (e.g. STORAGE_TYPE=s3)
     • set env S3_ACCESS_KEY to your Cloudflare R2 Access Key ID
     • set env S3_SECRET_KEY to your Cloudflare R2 Secret Access Key
     • set env S3_BUCKET to your Cloudflare R2 Bucket Name
     • set env S3_DOMAIN to your Cloudflare R2 Domain Name (e.g. https://<account-id>.r2.cloudflarestorage.com)
     • set env S3_DIRECT_URL_DOMAIN to your Cloudflare R2 Public URL Access Domain Name (Open Public URL Access, e.g. https://pub-xxx.r2.dev)

5. 📦 Min IO

   • Self Hosted
   • Reliable & Flexible Storage
   • Config (S3 Compatible):
     • set env STORAGE_TYPE to s3 (e.g. STORAGE_TYPE=s3)
     • set env S3_SIGN_VERSION to s3v4 (e.g. S3_SIGN_VERSION=s3v4)
     • set env S3_ACCESS_KEY to your Min IO Access Key ID
     • set env S3_SECRET_KEY to your Min IO Secret Access Key
     • set env S3_BUCKET to your Min IO Bucket Name
     • set env S3_DOMAIN to your Min IO Domain Name (e.g. https://oss.example.com)
     • [Optional] If you are using CDN, you can set S3_DIRECT_URL_DOMAIN to your Min IO Public URL Access Domain Name (e.g. https://cdn-hk.example.com)

6. ❤ Telegram CDN

   • Free Storage (Rate Limit)
   • Support Direct URL Access (China Mainland User Unfriendly)
   • Config:
     • set env STORAGE_TYPE to tg (e.g. STORAGE_TYPE=tg)
     • set env TG_ENDPOINT to your TG-STATE Endpoint (e.g. TG_ENDPOINT=https://tgstate.vercel.app)
     • [Optional] if you are using password authentication, you can set TG_PASSWORD to your TG-STATE Password

• OCR_ENABLED Image OCR Enabled (1 for Enabled, 0 for Disabled, Default is Disabled)
• OCR_ENDPOINT Paddle OCR Endpoint (Deploy PaddleOCR API)
  • e.g.: http://example.com:8000

Advanced OCR Config:

• OCR_SKIP_MODELS: Skip OCR Models List (Commonly for Vision Models)
  • e.g.: gpt-4-v,gpt-4-vision-preview,gpt-4-turbo, then the service will skip these models and directly store the image.
    • Tips: Each model has character inclusion matching, so when you set gpt-4-v model, it will skip all models that contain gpt-4-v (like azure-gpt-4-vision-preview, gpt-4-vision-preview will be also matched).
• OCR_SPEC_MODELS: Specific OCR Models List (Commonly for Non-Vision Models)
  • then although the image has marked as SKIP_MODELS, the service will still ocr process the image with this model first.
  • for example, when you set gpt-4-turbo to SKIP_MODELS (because gpt-4-turbo support vision and don't need to use OCR, gpt-4-turbo-preview cannot vision and need OCR), commonly the gpt-4-turbo-preview will be marked as gpt-4-turbo and skipped, then you can set gpt-4-turbo-preview to SPEC_MODELS to force OCR process.

EXAMPLE OCR Config:

OCR_ENABLED=1
OCR_ENDPOINT=http://example.com:8000
OCR_SKIP_MODELS=vision,gpt-4-v,gpt-4-all,gpt-4-vision-preview,gpt-4-1106-vision-preview,gpt-4-turbo,gemini-pro-vision,gemini-1.5-pro,claude-3,glm-4v
OCR_SPEC_MODELS=gpt-4-turbo-preview,claude-3-haiku

• ~/config.py: Env Config
• ~/main.py: Entry Point
• ~/utils.py: Utilities
• ~/handlers: File Handlers
• ~/store: Storage Handlers
• ~/static: Static Files (if using local storage)

• Python & FastAPI

Apache License 2.0