Quick Start

Three steps. Under two minutes. No SDK to install.

1. Get your API key

Sign up at your dashboard and create an API key. Keys start with comp_ and look like comp_xxxxxxxxxxxx.

You'll also add your LLM provider key (OpenAI, Anthropic, etc.) in the dashboard. Compresh needs it to forward requests on your behalf.

2. Point your client at Compresh

Change your base_url to https://api.compre.sh/v1. That's it.

from openai import OpenAI

client = OpenAI(
    api_key="comp_your_key",
    base_url="https://api.compre.sh/v1"
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"}
    ]
)

import OpenAI from "openai";

const client = new OpenAI({
    apiKey: "comp_your_key",
    baseURL: "https://api.compre.sh/v1"
});

const response = await client.chat.completions.create({
    model: "gpt-4o",
    messages: [
        { role: "system", content: "You are a helpful assistant." },
        { role: "user", content: "Hello!" }
    ]
});

curl https://api.compre.sh/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer comp_your_key" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hello!"}
    ]
  }'

3. Done

Seriously — that's it. Compresh proxies your requests, compresses context on multi-turn conversations, and returns responses in the exact same format your app already expects.

What works out of the box

• Streaming — SSE responses work identically
• Function calling & tools — tool definitions and calls are preserved
• All OpenAI-compatible models — GPT-4o, GPT-4o-mini, o1, o3, etc.
• System prompts — compressed separately with their own strategy

Zero code changes beyond the base_url.

Next steps

• Authentication — API key details, provider key setup, security
• Overview — How Episodic Memory Architecture works under the hood