A guide that travels with you
Your AI.
Your choice.
Bring the intelligence you already use to the things you’re exploring. You choose the model. You control the account.
No site-wide shared inference balanceBuild your own agent workflow
Learn the parts through a working graph: inspect state, change the rules, try a repair loop, and run the matching Python code.
Open Agent WorkshopReal LangGraph orchestration with a deterministic practice writer. No model calls or account needed.
Choose how to use AI
Change this anytimeUse the interactive Supply & Demand Lab inside your AI app →
Native Osiris is an invitation-only pilot. A running subscription service and your provider’s sign-in are required. The deployed MCP connector is a separate way to use tools inside an AI app.
Private runtime connection
Connect through OpenAI’s sign-in, then ask questions in the AI workspace or directly inside a configured project. This uses the Codex access and limits of your ChatGPT plan.
Private pilot: the separate connection service must be running before sign-in works. GitHub Pages serves this interface.
Your service operator provides a service address and pilot access code. No ChatGPT password or subscription token is entered on Ballzatram.
Set up the connection service ↗Keep using the account you have.
We prepare your question and selected context. Copy it, open your chat app, and paste it there. Your chat app’s plan limits apply. Answers stay in that app; they do not sync back automatically.
One connection. Your choice of models.
OpenRouter lets you use models from OpenAI, Anthropic, Google, and other providers. This uses your own OpenRouter balance; a ChatGPT or Claude subscription does not cover it.
The connection key stays in this tab’s session storage for up to four hours. Browser session restore may retain it until expiry. Disconnect to remove it from this tab; revoke it in OpenRouter to disable it everywhere.
The catalogue and connection check use no generation tokens.
Use an OpenRouter API key instead
Build and test without burning tokens.
This mode returns a fixed reading checklist using your question and context labels. It makes no external requests and does not generate AI analysis.
Advanced: use a direct OpenAI or Anthropic API key
Use a relay you operate or trust. Your API key and selected context pass through that relay to the chosen provider. Your provider bills your API account separately from your chat subscription.
The key is limited to this tab session and this exact relay/provider combination, for up to four hours. Changing the relay requires entering the key again.
On-site Osiris: connect a running runtime and your ChatGPT account to begin.
The Osiris workspace
What are you exploring?
Your workspace
Ready when you areYour model’s answer will appear here after you connect and explicitly send your question.
Paste the prepared prompt into the chat app and send it there. Opening the app does not send any data.