The universal token compressor that lets your wallet breathe (and your AI Agents too)

If you build with LLMs, you know the drama: agent loops that keep repeating the same context, giant JSON files clogging up the context window, and those end-of-month API bills that give you chills.

That's why I decided to launch Synthelion — a universal Token Compressor and Prompt Manager designed to cut token usage by up to 70%, speeding up model responses without losing a shred of meaning.

And yes, there's a quote I couldn't resist:

"Why use many tokens when few tokens do the same job?" — A Neanderthal man (and your wallet).

🔥 What is Synthelion and how does it work?

Synthelion sits between your code (or your agent) and the LLM. It analyzes the prompt, strips out all the superfluous "grammatical packaging" (articles, prepositions, conjunctions) and reduces words to their base form (lemmatization). The AI receives the exact same essential information, but distilled.

Before and after (A few practical examples)

1. Italian prose (~52% savings):
2. Before: I would like to know if it's possible to get information about cheap restaurants in Rome, please.
3. After: know possible get information cheap restaurant Rome
4. JSON Array (~69% savings):
5. Before: A classic JSON with repeated keys for every object.
6. After: Automatically converted into a clean, lossless Markdown table, which models digest natively with a fraction of the tokens.
7. HTML pages, build logs and Git diffs:
8. Synthelion includes a smart Content Router: it understands what you're passing it (code, logs, tables or text) and applies the perfect compression algorithm for that format.
9. 
   

🛠️ Universal Integration (Zero configuration)

Synthelion was built to be dropped in anywhere in 5 minutes:

1. 🔌 MCP Protocol: Native support for Claude Code, Cursor, Windsurf, and Claude Desktop. Just add one line to your config file to give your IDE compression superpowers.
2. 🐍 OpenAI & LangChain Plugins: Ready-made tools to pass as tools to the OpenAI API or into your ReAct agents on LangGraph.
3. 💻 CLI Interface and Python API: You can use it directly from the terminal in Bash pipelines or integrate it into your scripts with pip install synthelion.

🌍 Sustainability, by the numbers

It's not just about costs (which at volumes of 10M tokens a day mean thousands of dollars saved per year), but also about environmental impact.

Synthelion includes a built-in energy estimator: every token saved avoids about 0.005 mWh of compute energy and 0.002 mg of CO₂. You can track efficiency directly from your code:

Python

📦 Where to start?

The project is open-source, supports more than 50 languages out-of-the-box (auto-detected) and requires no AI model to run locally. It's the natural evolution of one of my earlier C# projects (Caveman), redesigned from scratch for the Python and AI Agents ecosystem.

1. 🐍 PyPI: pip install synthelion
2. 🛠️ GitHub Repository: francescopaolopassaro/synthelion