Vidai is the AI control plane for enterprise organisations that need to govern every LLM request inside their own perimeter. It is a single Rust binary that sits in the path of every AI call, between the applications making the calls and the model providers serving them. Drop it in by changing one base URL; any SDK, any provider.
From a single boundary, Vidai delivers per-team cost attribution, enforced spend controls, content-aware policy (PII redaction, region locks, model allow-lists), and audit-grade event logging mapped to the regulatory frameworks regulated buyers face today: EU AI Act, DORA, GDPR Article 44, UK CTPR, FCA, and ISO/IEC 42001. Audit evidence is recorded inside the customer's own infrastructure, with zero data egress.
Vidai runs cross-provider by design: OpenAI, Anthropic, Gemini, Bedrock, Vertex, Azure OpenAI, and self-hosted open-weight models (Llama, Mistral, your fine-tunes) are all governed from the same boundary. Cost-based routing sends each request to the cheapest model that meets the policy and quality bar; budget breakers stop the next call when a cap is reached, instead of sending an alert about a call that has already run.
Verified on a single legacy 8-core node: 21,803 requests per second sustained, 1.95 ms median overhead under load, 25 MB binary, zero data egress. Vidai is built for regulated mid-market and enterprise teams who cannot use SaaS AI gateways because prompts must not cross a vendor's perimeter.
