Stop paying for
tokens you don't need

import httpx

api_key = "tt_your_api_key"
text = "Your long prompt here..."

# Compress before sending to OpenAI/Claude/Gemini
resp = httpx.post(
    "https://tokenstree.es/compress",
    headers={"X-API-Key": api_key},
    json={"text": text, "target_token": 150},
)
compressed = resp.json()["compressed_text"]
savings = resp.json()["metrics"]["savings_percent"]
# → "83.2"  (saved 83% of tokens)

# Now use compressed in your LLM call
llm_response = your_llm_client.chat(compressed)

curl -X POST https://tokenstree.es/compress \
  -H "X-API-Key: tt_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{"text": "Your prompt here...", "target_token": 150}'

# Response:
{
  "compressed_text": "...",
  "metrics": {
    "tokens_original": 847,
    "tokens_compressed": 143,
    "tokens_saved": 704,
    "compression_ratio": "5.9x",
    "savings_percent": 83.1
  }
}

# Full pipeline: LLMLingua-2 + Tokinensis for maximum savings
resp = httpx.post(
    "https://tokenstree.es/compress/pipeline",
    headers={"X-API-Key": api_key},
    json={
        "text": text,
        "target_token": 200,
        "use_tokinensis": True,
        "tokinensis_version": 2,
        "translation_api_key": "your_tokentranslation_key",
    }
)
data = resp.json()
final = data["final_text"]
total = data["total_metrics"]["savings_percent"]
# → "94.7"  (saved 94.7% of tokens!)

# Save this to: ~/.claude/skills/tokenstransfer/SKILL.md

---
name: tokenstransfer
description: >
  Compress LLM prompts using TokensTransfer (LLMLingua-2).
  Activate when asked to compress, shorten or optimize a prompt
  before sending to an LLM. Reduces tokens by 50-91%.
---

# TokensTransfer Skill

You have access to the TokensTransfer compression API at tokenstree.es.
Use it to compress any text before it is sent to an LLM.

## API
POST https://tokenstree.es/compress
Header: X-API-Key: tt_YOUR_API_KEY
Body: {"text": "<prompt>", "rate": 0.3}

## When to use
- User asks to "compress", "shorten", "optimize" a prompt
- User wants to reduce LLM API costs
- Text is >50 tokens and will be sent to an LLM

## How to use
1. Take the user's prompt text
2. POST to /compress with rate=0.3 (keep 30% of tokens)
3. Return the compressed_text and savings_percent
4. Optionally pass to another LLM

## Rate guide
- rate=0.5 → gentle compression, 50-58% savings
- rate=0.3 → strong compression, 71-76% savings (recommended)
- rate=0.1 → aggressive, 89-91% savings (may lose detail)

# 1. Create skill directory
mkdir -p ~/.claude/skills/tokenstransfer

# 2. Save SKILL.md (see previous tab) to that directory
# 3. Add your API key to the SKILL.md file

# 4. In Claude Code, activate the skill:
/skill tokenstransfer

# 5. Then ask Claude to compress your prompt:
"Compress this prompt before sending to GPT-4: [your text]"

# Or use it directly via API in a pipeline:
import httpx, openai

compressed = httpx.post(
    "https://tokenstree.es/compress",
    headers={"X-API-Key": "tt_your_key"},
    json={"text": original_prompt, "rate": 0.3}
).json()["compressed_text"]

# Now use compressed instead of original_prompt
response = openai.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": compressed}]
)
# Saved 70%+ on tokens — same quality response