@relayplane/proxy

Intelligent AI model routing proxy for cost optimization and observability.

Installation

npm install -g @relayplane/proxy

Quick Start

# Set your API keys
export ANTHROPIC_API_KEY=your-key
export OPENAI_API_KEY=your-key

# Start the proxy
relayplane-proxy

# Configure your tools to use the proxy
export ANTHROPIC_BASE_URL=http://localhost:3001
export OPENAI_BASE_URL=http://localhost:3001

# Run your AI tools (Claude Code, Cursor, Aider, etc.)

Features

• Intelligent Routing: Routes requests to the optimal model based on task type
• Cost Tracking: Tracks and reports API costs across all providers
• Provider Agnostic: Works with Anthropic, OpenAI, Gemini, xAI, and more
• Local Learning: Learns from your usage patterns to improve routing
• Privacy First: Never sees your prompts or responses

CLI Options

relayplane-proxy [command] [options]

Commands:
  (default)              Start the proxy server
  telemetry [on|off|status]  Manage telemetry settings
  stats                  Show usage statistics
  config                 Show configuration

Options:
  --port <number>    Port to listen on (default: 3001)
  --host <string>    Host to bind to (default: 127.0.0.1)
  --offline          Disable all network calls except LLM endpoints
  --audit            Show telemetry payloads before sending
  -v, --verbose      Enable verbose logging
  -h, --help         Show this help message
  --version          Show version

Telemetry

RelayPlane collects anonymous telemetry to improve model routing. This data helps us understand usage patterns and optimize routing decisions.

What We Collect (Exact Schema)

{
  "device_id": "anon_8f3a...",
  "task_type": "code_review",
  "model": "claude-3-5-haiku",
  "tokens_in": 1847,
  "tokens_out": 423,
  "latency_ms": 2341,
  "success": true,
  "cost_usd": 0.02
}

Field Descriptions

| Field | Type | Description | |-------|------|-------------| | device_id | string | Anonymous random ID (not fingerprintable) | | task_type | string | Inferred from token patterns, NOT prompt content | | model | string | The model that handled the request | | tokens_in | number | Input token count | | tokens_out | number | Output token count | | latency_ms | number | Request latency in milliseconds | | success | boolean | Whether the request succeeded | | cost_usd | number | Estimated cost in USD |

Task Types

Task types are inferred from request characteristics (token counts, ratios, etc.) - never from prompt content:

• quick_task - Short input/output (< 500 tokens each)
• code_review - Medium-long input, medium output
• generation - High output/input ratio
• classification - Low output/input ratio, short output
• long_context - Input > 10,000 tokens
• content_generation - Output > 1,000 tokens
• tool_use - Request includes tool calls
• general - Default classification

What We NEVER Collect

• ❌ Your prompts
• ❌ Model responses
• ❌ File paths or contents
• ❌ Anything that could identify you or your project

Verification

You can verify exactly what data is collected:

# See telemetry payloads before they're sent
relayplane-proxy --audit

# Disable all telemetry transmission
relayplane-proxy --offline

# View the source code
# https://github.com/RelayPlane/proxy

Opt-Out

To disable telemetry completely:

relayplane-proxy telemetry off

To re-enable:

relayplane-proxy telemetry on

Check current status:

relayplane-proxy telemetry status

Configuration

Configuration is stored in ~/.relayplane/config.json.

Set API Key (Pro Features)

relayplane-proxy config set-key your-api-key

View Configuration

relayplane-proxy config

Usage Statistics

View your usage statistics:

relayplane-proxy stats

This shows:

• Total requests and cost
• Success rate
• Breakdown by model
• Breakdown by task type

Environment Variables

| Variable | Description | |----------|-------------| | ANTHROPIC_API_KEY | Anthropic API key | | OPENAI_API_KEY | OpenAI API key | | GEMINI_API_KEY | Google Gemini API key | | XAI_API_KEY | xAI/Grok API key | | MOONSHOT_API_KEY | Moonshot API key |

License

MIT