Press kit.

The major coding assistants — Claude Code, Cursor, Copilot — refuse, hedge, or rewrite into uselessness on a wide swath of work that's legitimate, billed, professional security engineering: exploit development, payload generation, reverse-engineering, evasion research, malware analysis, custom tradecraft, scrapers at scale. The reason isn't the engineer's intent. It's that the foundation labs above each tool wrote one content policy for the median web-app developer and applied it to everyone.

TartarusAI was built for the security professional underneath that policy. It runs a custom-tuned coding model on dedicated GPU infrastructure that we operate — no upstream content policy to inherit, no vendor three layers up who can change the rules mid-engagement. The agent ships with the runtime safety guards that actually matter for multi-file engagement work: a verification gate that runs your build before the model is allowed to say "done," read-before-overwrite, loop guards, failed-path blacklist.

Pricing spans four tiers: Lite at $20 (a 2-day unlimited trial), Pro at $250/month (80M tokens), Pro+ at $350/month (200M tokens, dedicated capacity), and Enterprise at $1,000/month (750M tokens, multi-seat, SSO, on-prem option). All paid in crypto — no card on file, no recurring charges. Refunds are tied to the product claim: if the agent refuses, hedges, or returns neutered output on legitimate engagement work, the unused portion is refunded — at any point in the subscription. Customers are red-team consultants, vulnerability researchers, exploit developers, malware analysts, and offensive-security labs. Privacy is documented in plain English: no training on customer prompts, 24-hour auto-purge of inactive sessions, NDA on file for Enterprise-tier accounts, on-prem deployment available.