Modulus AI Launches Mix, a Privacy-Focused Service That Forgets

Unlike standard AI tools that store queries for years, Mix treats every interaction as ephemeral. The service employs a bifurcated architecture where billing and reasoning occur on physically separate systems. A temporary authorization token allows a user to access AI capabilities without passing payment identity into the reasoning environment, ensuring that the model processes information without knowing the source.

Beyond privacy, Mix enhances output reliability by utilizing multiple frontier models, such as Claude Fable 5 and GPT-5.6 Sol Pro. Instead of relying on a single source, the service assigns independent reasoning roles to different models, forcing them to critique each other's conclusions and identify potential errors. Founder Richard Gardner notes that this multi-model approach mitigates the risk of a single AI sounding certain while being fundamentally incorrect.

The service operates on a prepaid credit model rather than a recurring subscription, charging $20 for 1,000,000 non-expiring tokens. Developed from Modulus AI’s decades of experience in high-performance computing and mission-critical software, the platform is now open for early access registration at mix.chat.