security-one-27b
Decision model for security
security-one-27b classifies prompts, code changes, logs, and alerts against the questions you define. It returns probabilities, so your application can handle clear cases and send uncertain events for deeper review.
how it works
Prompt-injection screening
security-one-27b scores prompts, retrieved text, and tool results for injection attempts. Prompt-injection detection is the release's best-validated security use case.
Security triage
security-one-27b scores the answers you define for code changes, logs, and alerts. Your application can use those probabilities to route events to analysis or review.
Hosted or self-hosted inference
Superagent hosts inference through the API. The same Apache 2.0 model weights can run on your infrastructure.
use cases
Supabase log triage
security-one-27b scores Supabase log batches. The connected workflow uses those probabilities to route activity to monitoring, investigation, or human review, with thresholds evaluated on your traffic.
See the Supabase example →Screening before agent delivery
security-one-27b scores untrusted input for prompt injection. The connected application handles clear results under its policy and sends uncertain inputs for deeper review.
See the screening example →FAQ
more reading
Try a security decision
Send a question to security-one-27b and connect its probabilities to your application's response policy.