The world’s first AI that doesn’t say No.

What Ghost is for

Ghost is a hosted AI from The Sapient Directive. It is built for authorized technical work that heavily safety-tuned assistants often decline: security research and red-teaming, AI/ML evaluation, synthetic data generation, and deep technical research across security, ML and science.

You use it the same way you use other modern assistants: in a web chat, through Claude Code / the Anthropic Messages API, or through an OpenAI-compatible chat completions endpoint.

Models and surfaces

Ghost hosts open-weight models offered as Ghost X and Ghost X1. Pro and Max include both, on web, terminal and API, with agent tools (web search + code) and website replication.

The Connect screen in the app gives you a Claude Code command that points ANTHROPIC_BASE_URL at the Ghost gateway with your own key.

Privacy by construction

Ghost is designed for zero conversation retention. Prompts and outputs are not stored, logged or used for training. Each session runs in an isolated GPU container that is destroyed on exit.

Pricing

Free — try Ghost, no card needed: 3 messages to Ghost X1 and a website replication.

Pro — $20 a month. Unlimited messaging and coding for everyday use; Ghost X and Ghost X1 on web, terminal and API; agent tools; 10 website replications a month; 5-hour session usage window.

Max — from $50 a month. Everything in Pro with 2.5x more usage (5x and 20x steps available), unlimited website replications, higher output limits, early access and priority access.

Cancel anytime. Start free at https://thesapientdirective.com/ghost/signup.

Safety boundary

Ghost is low-refusal, not unlimited. A fixed safety layer reviews every request and declines a narrow, clearly defined set of bright-line categories. No prompt, system message or setting can move that boundary. Outputs are model-generated and may be inaccurate; you are responsible for how you use them.

FAQ

What is Ghost?

Ghost is a hosted, low-refusal AI model from The Sapient Directive for authorized technical work that heavily safety-tuned assistants often decline: security research, AI red-teaming, evaluation and synthetic-data generation, and deep technical research. It runs open-weight models (Ghost X and Ghost X1) and is available in the browser, in the terminal and over an API.

How much does Ghost cost?

Ghost has a Free plan (3 messages to Ghost X1, no card needed), Pro at $20 a month (unlimited messaging and coding for everyday use) and Max from $50 a month (2.5x more usage than Pro, with 5x and 20x steps). You can cancel anytime.

Can I use Ghost with Claude Code or my existing SDK?

Yes. Ghost exposes an Anthropic Messages-compatible endpoint and an OpenAI-compatible chat completions endpoint. The Connect screen in the app gives you a Claude Code command that points ANTHROPIC_BASE_URL at the Ghost gateway with your own key; for the OpenAI SDK you swap the base URL and key.

Does Ghost store my prompts?

Ghost is built for zero conversation retention: prompts and outputs are not stored, logged or used for training, and each session runs in an isolated GPU container that is destroyed on exit. See the privacy policy for the full terms.

What will Ghost not do?

Ghost has a fixed safety boundary. A safety layer reviews every request and declines a narrow, clearly defined set of bright-line categories. No prompt, system message or setting can move that boundary.

Is Ghost an uncensored AI?

Ghost is a low-refusal model: it engages directly with legitimate security, research and data-generation requests that other models often refuse out of caution, but it is not unlimited. A fixed, narrow safety boundary still applies, and you are responsible for how you use the outputs.

Which models does Ghost run?

Ghost hosts open-weight models, offered as Ghost X and Ghost X1. Pro and Max include both, on web, terminal and API.