Skip to content

Installation

This guide covers installing CheckStream on your system.

Prerequisites

  • Rust 1.70+ (for building from source)
  • Git (for cloning the repository)
  • 8GB+ RAM recommended for ML models

Building from Source

1. Clone the Repository

git clone https://github.com/Skelf-Research/checkstream
cd checkstream

2. Build with ML Support

For full ML classifier support (recommended):

cargo build --release --features ml-models

For pattern-only classifiers (smaller binary, faster build):

cargo build --release

3. Download Models

Download pre-trained models from HuggingFace:

./scripts/download_models.sh

This downloads: - DistilBERT sentiment classifier - Toxicity detection model - Prompt injection detector

Models are cached in ~/.cache/huggingface/.

Docker Installation

Pull and run the official Docker image:

docker pull checkstream/checkstream:latest

docker run -d \
  -p 8080:8080 \
  -v $(pwd)/config.yaml:/app/config.yaml \
  -v $(pwd)/policies:/app/policies \
  checkstream/checkstream:latest

Docker Compose

version: '3.8'
services:
  checkstream:
    image: checkstream/checkstream:latest
    ports:
      - "8080:8080"
      - "9090:9090"  # Metrics
    volumes:
      - ./config.yaml:/app/config.yaml
      - ./policies:/app/policies
    environment:
      - RUST_LOG=info
      - CHECKSTREAM_BACKEND_URL=https://api.openai.com/v1

Verifying Installation

Check that CheckStream is running:

# Health check
curl http://localhost:8080/health

# Expected response:
# {"status":"healthy","version":"0.1.0"}

Check readiness (ensures classifiers are loaded):

curl http://localhost:8080/health/ready

Configuration Files

CheckStream uses YAML configuration files:

File Purpose
config.yaml Main proxy configuration
classifiers.yaml ML model definitions
policies/*.yaml Safety policies

See Configuration for detailed options.

Next Steps