73% Less Waste.
Up to 10× More Capable.
THERION Protocol is a set of instruction files that give any AI a better personality. Drop them into any AI-powered tool that reads files — Claude Code, GitHub Copilot, Cursor, Windsurf, and more. The moment your AI reads them, its behavior changes: it stops hedging, stops asking permission, stops delivering fragments. It routes to the right specialist automatically, delivers complete work, and executes without being asked twice. Research, writing, planning, analysis, code, automation — any domain, any task. Nothing to install. Three words to activate. Free forever.
Measured in tokens. Proven in production.
Every AI coding tool charges by tokens or time. Here's exactly where those resources go — and how THERION reclaims them.
The average non-trivial AI task wastes ~7,000 tokens on preambles, permission loops, and follow-up rounds. THERION's 11 Iron Laws eliminate those defaults, cutting total task cost to ~2,300 tokens.
Multi-step tasks with fragment follow-ups can burn 4,000–11,100 tokens of waste. THERION's phased execution and todo discipline collapse that to near zero — one structured prompt, specialist routing, complete output.
3–5 minutes saved per task. At 25 tasks/day: 75–125 minutes recovered. At a $75/hr loaded cost, that's $2,000–3,500/mo per developer in recaptured productivity.
Works with Claude, Gemini, Mistral, Llama, and any capable LLM. Runs in VS Code, Cursor, Windsurf, Claude Code, and more. Architecture-driven, not model-locked.
No API keys, no subscriptions, no accounts. Drop one folder into your project. The protocol is instruction files — plain text that works everywhere. AGPL v3.
Zero-config agent routing.
THERION scans every message for domain keywords and automatically routes to the right specialist. You never select an agent — THERION decides in milliseconds. Your message flows through: keyword detection → domain match → specialist loaded → execution.
No existing agent? THERION builds one.
When your task spans multiple domains or doesn't map to any existing specialist, THERION synthesizes a hybrid agent on-the-fly from the 2–3 closest domains. It states the hybrid in one line and executes immediately. No delay. No permission needed. 67 specialists cover the common ground — synthesis covers the frontier.
Three steps. Under a minute.
No technical knowledge required. If you can open a folder and type in a chat box, you can use THERION.
Download the Files
Go to github.com/erevusobolus/THERION-SYSTEM and download the ZIP, or clone it if you know how. You'll get a folder with a few text files inside.
Open in Your Code Editor
Open the folder in VS Code, Cursor, Windsurf, or any editor with AI chat. The instruction files are picked up automatically — no configuration needed.
Say the Magic Words
Type "WAKE UP THERION" in your AI chat. The protocol activates instantly. Your AI now behaves differently — more capable, more autonomous, more useful.
Clone repository → Open folder → Type "WAKE UP THERION" → Upgraded. Zero accounts. Zero config. Zero cost. Any AI.
Four small files. That's all it takes.
THERION reads four text files when it starts up. Each one teaches the AI something different about how to behave. You can customize them to match your preferences.
The personality file. This is what makes the AI stop being cautious and start being useful. It defines how the AI should respond and act.
The expert directory. This file tells the AI about all 67 specialist modes across 12 domains and when to switch between them automatically.
Your preferences. Tell the AI what languages you use, what frameworks you prefer, and how you like your code. It remembers.
The memory file. Lessons from previous sessions — mistakes to avoid, things that worked well. The AI gets smarter over time.
67 expert modes. Auto-detected.
Instead of one generic AI, THERION gives you 67 specialized experts across 12 domains. Ask about design and it becomes a frontend specialist. Ask about servers and it becomes a backend expert. It switches automatically based on your question — no commands needed.
11 Iron Laws your AI will follow.
These 11 laws eliminate the most costly AI behaviors: incomplete code, permission loops, excessive caveats, and fragments that require follow-up rounds. Once THERION is active, your AI enforces them automatically — on every prompt, every task, every output.
Built for Claude Code. Works everywhere else too.
Claude Code is the most popular AI coding tool right now — and THERION was built with it in mind. But the protocol works with any AI that reads instruction files: GitHub Copilot, Cursor, Windsurf, VS Code + Claude, VSCodium. If your AI can read a folder, THERION upgrades it.
Your AI is waiting for this.
Three files. Any AI. Any task. Clone the repository, type three words, and start receiving complete work — not prompts that need more prompts. Claude Code, Copilot, Cursor, Windsurf. Free. Forever.