Commitments
Building AI that is open where it matters and safe by architecture.
Four foundational commitments guide every model we train, every agent we orchestrate, and every system we operate.
Open AI for Everyone
Advanced AI should not leave regional and African languages behind. We release open-source models and benchmark tools so builders everywhere can participate.
- Free, MIT-licensed voice and language models like Ẹtí
- Honest reporting of both strengths and areas we are working to improve
- Lightweight builds so models run on standard computers without expensive cloud servers
Dependable Safety & Oversight
AI systems taking actions in the real world require practical guardrails, human approvals, and continuous monitoring.
- Clear permission boundaries before any external action is taken
- Real-time anomaly detection and complete, transparent audit logs
- Automatic fallbacks and recovery so operations remain stable
Data Privacy & Ownership
Organizations must retain complete ownership of their private knowledge, customer conversations, and internal data.
- Strict data separation so no customer information is ever mixed or shared
- Custom adaptations and fine-tuned models remain 100% your property
- Zero training on your private operational logs or customer chats
Honest, Real-World Benchmarks
We test our systems on real-world data and openly share results and limitations so teams know exactly what to expect.
- Openly published test scripts alongside our model releases
- Clear, honest benchmarks comparing clear speech with casual everyday conversation
- Continuous quality checks to ensure accuracy remains consistent over time
Read our governance & usage policy
Detailed standards on responsible deployment, dataset governance, and security disclosure.