A Scotland-based startup building a high-performance AI gateway with built-in guardrails for production LLM deployments.
We focus on what matters in production: security, reliability, latency, and control—not just swapping model providers.
⚙️ Vidai.Server (Enterprise)
An AI gateway that enforces runtime guardrails, rate limits, audit trails, and structured output guarantees—without sacrificing performance.
- Sub-15ms p99 latency, 6,000+ RPS per core
- Zero-copy Rust data plane
- Full sovereignty: your prompts, responses, and logs never leave your infrastructure
- Pluggable guardrail policies (e.g., PII redaction, prompt injection detection, schema enforcement)
Guardrail models (e.g., classifiers, validators) are optional extensions—you bring your own or use lightweight defaults.
Realistic, zero-dependency mock server for LLM APIs (OpenAI, Anthropic, Bedrock, etc.).
- Simulates streaming, tool calls, errors, and latency
- ~7MB static binary—no Docker, no DB
- Ideal for testing, CI, and local dev
🧰 VidaiSDK
Python SDK with drop-in OpenAI compatibility and automatic structured output.
- Enforces JSON/schema compliance at the client
- Works standalone or with Vidai.Server for end-to-end guarantees
- Supports aliasing and clean provider abstraction
- Models can be optimized via ONNX or Candle for lower latency—but this is optional. The gateway works with any standard LLM endpoint.
- No vendor lock-in. No hidden telemetry.
- Built for teams that need production-grade control without bloated middleware.
🏴 Based in Scotland. Open source where it helps developers. Enterprise-ready where it matters.