Unrestricted AI for everyone.
Uncensored intelligence, for anyone. The model that never says no to anything.
It exploits the vulnerability, not just reports it.
One prompt, then the agent works the target: confirm the injection, exploit it to pull the data it exposes, and drop a working proof-of-concept. Every step is a tool call, the ones that touch a shell or a file ask first, and every session is a transcript you can read back.
Built for a repository you don't trust.
Our own agent harness: 40 tools, five modes that narrow what each turn sees, and defaults for a repository you have not audited.
A hostile repository cannot turn a read into execution.
Every git command carries ten neutralising -c flags, closing the hooks, diff drivers and transports a repository can trigger from its own config. Reading an untrusted branch stays a read.
Permissions are gates, not suggestions.
Tools declare whether they write, shell out, or reach the network - all false in a fresh config. Hiding a tool from a mode is not a gate; called by name, it still hits one.
The request is measured before it is sent.
All 40 tools serialise to 24,323 characters - about 6,000 tokens on every request. Each mode advertises only what it needs, and the agent counts the request before sending.
Ultrac is for teams that need controlled access.
Governments, organizations, and individuals use Ultrac for authorized offensive and defensive security work - and the code around it.
Governments
Dedicated deployments for classified or regulated work in your jurisdiction.
- Fuse drone, satellite, and signals feeds into one intelligence picture
- Accelerate drug discovery and research
- Offensive and defensive cyber operations
- Screen travelers and cargo at the border
Organizations
API access for security analysis, incident response, and code review.
- Find an attacker hiding in your logs
- Break down malware and how to block it
- Break into your own systems to find the holes
- Scan your whole codebase for security bugs
Individuals
Security research and coding workflows for individual operators.
- A model that never answers with "I can't help with that"
- Write and explain exploit code for your own machines
- Take apart programs and solve hacking challenges
- Run focused sessions from the dashboard or CLI
Questions
Access is issued by key, logged, and revocable. Sovereign deployments can be configured with zero data retention.
Ultrac answers the authorized security and coding questions consumer models over-refuse. The hard out-of-scope lines are on the Security page.
Ultrac runs in our cloud. Use the shared API, a dedicated in-region instance, or a Sovereign deployment in your jurisdiction.
Yes. Point an OpenAI SDK at the Ultrac base URL and use the cosmic-1 model ID.
The API is invite only. Request access for an early key.
Bring Ultrac into the work.
Run the agent in your terminal, or call the same model from chat or an OpenAI-compatible API, when research has to turn into a patch.