Edgee AI Adapter for Paperclip

A Paperclip adapter that routes all LLM requests through Edgee AI Gateway for token compression and cost savings.

Key Features

• Token Compression — Reduces token usage by 10-70% through intelligent compression
• Cost Savings — Lower API costs while maintaining response quality
• Multi-Provider Support — Works with Anthropic, OpenAI, Google, Meta, Mistral models
• Usage Tracking — Built-in tracking of tokens saved and cost reductions

Installation

Install from npm

npm install @tecknix/edgee-paperclip-adapter

Quick Start

1. Create an Edgee agent in Paperclip

In the Paperclip UI, create an agent with adapter type edgee_ai:

{
  "name": "Edgee AI",
  "adapterType": "edgee_ai",
  "adapterConfig": {
    "edgeeApiKey": "your-edgee-api-key",
    "edgeeModel": "anthropic/claude-sonnet-4-5"
  }
}

2. Assign work

Create issues in Paperclip and assign them to your Edgee-enabled agent. Each run will automatically route through Edgee for compression.

Configuration Reference

Required Fields

| Field | Type | Description | |-------|------|-------------| | edgeeApiKey | string | Your Edgee API key |

Optional Fields

| Field | Type | Default | Description | |-------|------|---------|-------------| | edgeeModel | string | anthropic/claude-sonnet-4-5 | Model to use via Edgee |

Supported Models

| Provider | Model ID | Label | |----------|----------|-------| | Anthropic | anthropic/claude-haiku-4-5 | Claude Haiku (Fast) | | Anthropic | anthropic/claude-sonnet-4-5 | Claude Sonnet (Balanced) | | Anthropic | anthropic/claude-opus-4-5 | Claude Opus (Power) | | OpenAI | openai/gpt-4o | GPT-4o | | OpenAI | openai/gpt-4o-mini | GPT-4o Mini | | Google | google/gemma-2-9b | Gemma 2 9B | | Meta | meta/llama-3.1-70b | Llama 3.1 70B | | Meta | meta/llama-3.1-8b | Llama 3.1 8B | | Mistral | mistral/mistral-large | Mistral Large | | Mistral | mistral/mistral-small | Mistral Small |

Development

Prerequisites

• Node.js 20+
• Paperclip server for local testing
• Edgee API key from edgee.ai

Clone and Setup

git clone https://github.com/TeckniX/edgee-paperclip-adapter
cd edgee-paperclip-adapter
npm install

Type Check

npm run typecheck

Test

npm run test

Architecture

Paperclip                     Edgee                    LLM Provider
    │                           │                          │
    ├─ Heartbeat fires ────────▶│                          │
    │                           │                          │
    │                           ├─ Compress prompt ──────▶│
    │                           │                          │
    │                           │◀─ Compressed response ───┤
    │                           │                          │
    │◀─ Return results ────────┤                          │
    │                           │                          │

1. Paperclip calls execute() with the prompt
2. Adapter routes the request through Edgee AI Gateway
3. Edgee compresses the prompt and response
4. Compressed response returns with usage stats

License

MIT