Skip to content

docker.md

Latest commit

 

History

History
102 lines (71 loc) · 2.32 KB

docker.md

File metadata and controls

102 lines (71 loc) · 2.32 KB

Docker Setup

Run the full TierFlow stack (router + ML classifier) with Docker Compose.

Quick Start

# Clone both repos
git clone <tierflow-repo> tierflow
git clone <llmrouter-service-repo> llmrouter-service

# Start everything
cd tierflow
docker compose up -d

# Check status
docker compose ps
docker compose logs -f

TierFlow will be available at http://localhost:18800.

Architecture

                   docker network
┌──────────────┐      ┌────────────────┐
│  tierflow  │─────▶│  llmrouter     │
│  :18800      │      │  :18801        │
│  (Node.js)   │      │  (Python)      │
└──────────────┘      └────────────────┘
  • tierflow waits for llmrouter to be healthy before starting
  • They communicate via http://llmrouter:18801/classify (Docker DNS)
  • Both bind to 127.0.0.1 on the host (localhost only)

Configuration

API Keys

Pass API keys as environment variables:

# Option 1: .env file
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
docker compose up -d

# Option 2: inline
ANTHROPIC_API_KEY=sk-ant-... docker compose up -d

Custom Config

Mount your config file:

# Default: mounts ./tierflow.config.json
docker compose up -d

# Custom path:
TIERFLOW_CONFIG=/path/to/config.json docker compose up -d

Without ML Classifier

If you only want the router (rule-based routing):

docker compose up tierflow -d

TierFlow falls back to the 14-dimension keyword scorer when the ML classifier is unavailable.

Image Sizes

Image Size Why
tierflow ~180MB Node.js slim + built JS
tierflow-llmrouter ~600MB Python + sentence-transformers model (80MB) baked in

The ML model is baked into the image to avoid downloading at runtime.

Health Checks

# Router health
curl http://localhost:18800/health

# ML classifier health
curl http://localhost:18801/health

# Dashboard
open http://localhost:18800/dashboard

Rebuilding

docker compose build --no-cache
docker compose up -d