Cosmic 1
Cosmic 1 is the model the Ultrac agent calls. It is set up for security and coding work and served where your work is allowed to run. The agent around it is what makes it useful.
What Cosmic 1 is.
An open-weights model, served for security and coding work behind an OpenAI-compatible API.
- Model ID
- cosmic-1
- Context window
- 262,144 tokens
- Reply reserve
- 4,096 tokens
- API format
- OpenAI-compatible chat completions
- Availability
- Invite only
Open weights, so it can run where you need it.
Open weights can be deployed inside a jurisdiction that will never allow a call to someone else's cloud, and swapped when a better model appears. Those are the two properties this product needed.
Security analysis
Analyze malware behavior, incident timelines, suspicious commands, vulnerabilities, and detection logic for systems you are authorized to protect or test.
Code review
Review code, explain risky paths, identify security bugs, and draft patches with enough context to preserve local conventions.
Tool calling
The model returns tool calls rather than prose, so the agent can use it to run commands, edit files, and work a multi-step investigation.
Controlled access
Access is issued by API key. Usage is scoped, logged, and revocable. Sovereign deployments add isolated hosting and zero-data-retention options.