The Critical AI platform.
Five products on one foundation: energy-based research models, knowledge that cites its sources, operations that run exactly once, defense that records the reason for every block, and intelligence that holds up as evidence.
One record of every decision.
Every Qriton product writes to the same decision record: the inputs, the model version, the reviewer, the action and the outcome.
When someone asks why, you open the record instead of reconstructing the case from logs. Use one product for one job, or connect them: because they share deployment, oversight and records, adding a second product is configuration, not a new project.
Two ways in. One model layer.
Developers work in the terminal with Qriton CLI. Operations run as Qriton Flow workflows, with Qriton KB as their source. Both use the same models, local first.
One foundation.
Decision records
Inputs, model version, reviewer and outcome kept together, reviewable and exportable.
Human oversight
Review rules and approvals route low-confidence or consequential cases to a person.
Your perimeter
Outbound connections only where your policy allows, and every one recorded.
Model choice
Open-source or approved cloud models, per workflow, with fallback.
Runs where your data lives.
- On-premises
- Your servers or datacentre. Platform, models and records stay inside your network, reachable only through your gateway.
- Private cloud
- Your tenant, your region, your keys. Data at rest stays in your storage under your own encryption.
- Air-gapped
- With a private model, a complete workflow runs end to end without a single external connection.
- Edge
- Qriton Shield runs on Windows, Linux, Docker and ARM64.
Your model strategy, not your vendor’s.
- Qriton HLM family
- Research models for language, vision, 3D, audio and sensor data, used alongside other models.
- Open-source models
- Your selection, on your infrastructure. Air-gap ready.
- Commercial providers
- OpenAI, Anthropic or Azure AI, where your policy allows. Metered.