Vidai - The AI Control Plane

Hey Product Hunt 👋 I'm Nagu, co-founder of Vidai, based in Scotland. My background is infrastructure where every millisecond and every byte cost real money: low-latency adtech, low-latency video streaming, and most recently AI/creative infrastructure at hyperscaler regulatory scale. Performance and regulatory discipline are the same instinct applied to different boundaries. That's the lens we built Vidai through. The pattern we kept seeing: a team picks one LLM provider, then a second, then a third. Within months the AI traffic is a sprawl, costs are exploding (agent workflows multiply per-task call counts 10× to 100×, and the dashboards built for chat-pace traffic don't update fast enough to see the runaway), and a compliance officer is asking a question the CSV export can't answer. To survive it, platform teams reach for LiteLLM for translation, something for traces, custom code for cost, custom code for guardrails, custom code for fallback. Seven libraries, seven CVE surfaces, seven on-call pages. Nobody built that stack on purpose. The middleware became a product you didn't mean to build. A good gateway gets a request to a model. That's a fine starting point. What gateways don't answer, and what surprised people most, is the question "do we have to send prompt content out to a SaaS just to make the routing decision?" Vidai extends the same boundary a gateway draws. One Rust binary doing all seven jobs in one pass, with bidirectional native translation built into the proxy path: call with the Anthropic SDK, route to OpenAI, and the response comes back in Anthropic shape. Native in, native out. No forced rewrite to an OpenAI- shaped client. Cost attribution, inline policy and full audit on the same request, on your perimeter. Provider-agnostic, framework-agnostic, no phone-home. We set ourselves four targets to make that work: hyperscaler-grade efficiency, ultra-low latency that holds at scale, extreme throughput on modest hardware, and drop-in integration that isn't a rewrite. We chose Rust because it lets us hit those bars together; the engineering decisions that got us to 1.95 ms median with cost, policy and audit active on a legacy 8-core node are ours. The full benchmark, the methodology, and the 3,400× p95-latency gap vs Python orchestration at 12k RPS are in the engineering blog: https://vidai.uk/blog/rust-pytho.... Two people in the business notice the difference. The CFO gets a live per-team / per-app / per-model spend view with budgets that actually stop the next call when breached. The CCO gets an audit trail from day one. You don't set up an audit trail the week the regulator emails; by then the evidence you need is the requests you ran six months ago. Vidai writes that trail from the first request, structured so it maps straight to ISO/IEC 42001 SoA, EU AI Act Article 12, DORA, GDPR Article 44. A few things to look at: The free Community edition, the full self-hosted control plane, runs anywhere with a CPU. Your VPC, your laptop, a Raspberry Pi if you really want to. Start at https://vidai.uk. The Rust-vs-Python performance write-up: https://vidai.uk/blog/rust-pytho... The comparison vs LiteLLM, Portkey, Kong AI, Bedrock: https://vidai.uk/ai-gateway-comp... The benchmark methodology (it's reproducible): https://vidai.uk/performance And Vidai.Mock, our open-source companion: a wire-accurate LLM mock server for CI. Apache 2.0, deterministic, no real LLM keys needed. Mock OpenAI, Anthropic, Gemini and Bedrock offline with production-accurate streaming and error behaviour: https://github.com/vidaiUK/Vidai... What I'd genuinely love feedback on: For teams already running a gateway: what was the moment a new question arrived that your gateway wasn't shaped to answer? (a regulator asking for evidence? agent-pace cost attribution? a second team's policy not matching the first?) Is the "Community is the full product" model the right call, or should we hold back features? We've made the bet that the self-hosted Community edition needs to be the production product for people to trust it for the audit path. Happy to answer anything. Technical, commercial, naming, whatever. If you're an engineer kicking the tyres tonight, drop the rough edges in here too. That's the most useful kind of comment we can get. Nagu 🏴󠁧󠁢󠁳󠁣󠁴󠁿 Scotland · https://vidai.uk