Swap base_url, cut LLM token costs 40-60%.
AgentReady is an OpenAI-compatible proxy. It compresses prompts using a deterministic rule-based engine (no secondary LLM call, ~5ms overhead) before forwarding to OpenAI / Anthropic / Google. Works with any OpenAI SDK.
Before / After — real system prompt
You are a helpful assistant. It is important to note that the user is asking about the process of photosynthesis. Essentially, photosynthesis is basically the method by which plants are able to convert sunlight into energy that they can use. In order to accomplish this, they utilize chlorophyll, which is a green pigment that is found in the leaves of the plants. The fact of the matter is that this process is fundamentally crucial for life on Earth as we know it, given that it is the primary source of oxygen in the atmosphere. Furthermore, it should be noted that without photosynthesis, the vast majority of life forms on this planet would not be able to survive. Additionally, the process also plays a very significant role in the carbon cycle, as it essentially removes carbon dioxide from the atmosphere and converts it into organic compounds. Please explain this topic clearly and concisely to the user.
You are a helpful assistant. The user asks about photosynthesis. Photosynthesis is the method plants convert sunlight into usable energy. They use chlorophyll, a green pigment in leaves. This process is crucial for life on Earth as the primary source of atmospheric oxygen. Without photosynthesis, most life forms would not survive. It plays a significant role in the carbon cycle, removing CO2 from the atmosphere and converting it into organic compounds. Explain this topic clearly and concisely to the user.
Same meaning, same model output. 102 tokens saved per call. At GPT-4o rates ($2.50/1M input), that adds up fast.
How it works
Your App ──▸ AgentReady Proxy ──▸ OpenAI / Anthropic / Google
│ │
│ Compresses prompts: │ Receives compressed
│ • filler words removed │ prompt, responds
│ • verbose → concise │ normally
│ • redundancy stripped │
│ • ~5ms, no LLM call │
│ │
10,000 tok ≈ 4,200 tokWhat it removes
- • Filler words (basically, essentially, it should be noted…)
- • Verbose phrases ("in order to" → "to")
- • Redundant connectors (however, furthermore…)
- • Duplicate lines & boilerplate
- • Excess whitespace
What it preserves
- • All semantic meaning (not a summarizer)
- • Code blocks & inline code
- • URLs, numbers, dates
- • Sentence structure & reading order
- • Multi-language (EN, IT, FR, DE, ES)
Integration — 30 seconds
pip install agentready-sdk && agentready initnpx agentready-sdk init · or grab a key at /quick-keyfrom openai import OpenAI
client = OpenAI(
base_url="https://agentready.cloud/v1", # ← only change
api_key="ak_...", # AgentReady key
default_headers={
"X-Upstream-API-Key": "sk-..." # your OpenAI key
}
)
# Every call is now compressed automatically
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": your_long_prompt}]
)Or monkey-patch existing code (zero changes):
# Zero changes to existing code
from agentready import patch_openai
patch_openai(api_key="ak_...")
# All existing OpenAI calls now go through AgentReady automaticallyCompression levels
| Level | Savings | Latency |
|---|---|---|
| light | 10-20% | ~2ms |
| standard ← default | 20-40% | ~5ms |
| aggressive | 35-55% | ~8ms |
FAQ
Why not just write better prompts?
You should. This handles the mechanical noise that even good prompts contain — articles, filler words, verbose constructions. Think of it as gzip for natural language.
Does compression hurt output quality?
Tested across GPT-4, Claude, Gemini with varied prompt types. Output quality delta is <2% on BLEU/ROUGE scores. The removed content is linguistically redundant — it never carried information the model needed.
Why should I trust a proxy with my API keys?
Your upstream key (sk-...) is passed via header and forwarded directly to the provider. We never store it, never log it. For zero-trust setups, use the SDK which compresses client-side before calling OpenAI directly.
What does it cost?
Free during beta. No credit card required. After beta: pay-per-token, still significantly less than the tokens you save.
OpenClaw Integration
Using OpenClaw? Add AgentReady in 1 minute.
OpenClaw is a self-hosted gateway that connects chat apps to AI agents. Add AgentReady as a custom provider in your openclaw.json and every agent call gets automatically compressed — zero code changes.
// ~/.openclaw/openclaw.json
{
"models": {
"providers": {
"agentready": {
"baseUrl": "https://agentready.cloud/v1",
"apiKey": "ak_your_key",
"headers": {
"X-Upstream-API-Key": "sk-your-openai-key"
}
}
}
}
}Works with
Try it in 30 seconds
Free during beta. No credit card.