FAQ
Questions, answered straight.
The things buyers actually ask. If yours isn't here, book a 20-minute demo and ask us.
Compatibility
Does it work with Azure OpenAI and Microsoft Copilot?
Yes for Azure OpenAI Service, Vidai sits in front of it like any other provider, with the same identity, routing and guardrails. Microsoft 365 Copilot itself (the in-app assistant) has no API and is licensed per user by Microsoft, so no gateway can sit in front of it; teams that want Copilot-class models programmatically use Azure OpenAI, which we fully support. Custom Copilot Studio agents can be integrated on request.
Which AI providers does Vidai support?
OpenAI, Azure OpenAI, Anthropic, Amazon Bedrock, Google Gemini and Vertex AI, Mistral AI, IBM watsonx and others, and your own models on Ollama or any OpenAI-compatible server. Adding or switching a provider is a configuration change, not a rebuild.
Do we have to rewrite our code to OpenAI format like other gateways?
No, and this is a real difference. Most gateways only accept the OpenAI dialect, so an app built on the Anthropic SDK or Google ADK has to be rewritten to use them. Vidai translates in both directions: your code keeps speaking the Anthropic SDK, Google ADK or Gemini natively, and we convert to and from whatever the target model speaks. The change is a one-line endpoint update, not a refactor.
Can different teams use different SDKs through one Vidai?
Yes. One team can stay on the Anthropic SDK, another on the Google ADK, another on OpenAI, all through the same Vidai, each in its own native dialect. Vidai is the common layer underneath, not a lowest-common-denominator everyone has to adopt.
Does it work with LangChain, LlamaIndex and agent frameworks?
Yes. Common orchestration and agent frameworks work through Vidai unchanged, including LangChain, LangGraph and the Google ADK.
How Vidai compares
How is Vidai different from security platforms like CrowdStrike?
They protect different places, and most organisations run both. Security platforms such as CrowdStrike are strong at threat detection on people's devices, browsers and SaaS apps, and now protect prompts and agents there too. Vidai sits in the path of every AI call where it goes to a model, including apps and agents running on servers that never touch a laptop, and governs how your organisation uses AI: who may call which model, what data may go, where it may run, what it costs, and the evidence an auditor needs. When Vidai blocks or redirects a call, the event streams to the SIEM your security team already watches, never containing prompts or responses.
How is Vidai different from Microsoft Purview?
Purview governs AI inside Microsoft 365, and it's strong there: Copilot governance, eDiscovery, retention and legal hold, sensitivity labels and Insider Risk. Vidai governs AI calls from any app, agent, model or cloud, not only Microsoft's, and does it inside your own network, so prompts and responses stay with you. It also does what a compliance suite doesn't set out to do: routing between models, cost per agent and application, and budgets that act. On Azure the two sit side by side, with Vidai's events going to Microsoft Sentinel or any SIEM.
We already have DLP and a secure web gateway such as Zscaler or Netskope. Do we still need Vidai?
Keep them. Secure access and DLP tools are strong at people using AI in a browser, at the edge of your network. Vidai covers what leaves from your servers: apps and agents calling models directly, what each prompt contains, and which tools an agent was offered. It also routes, prices and records every call, which an inspection layer doesn't set out to do.
How is Vidai different from guardrail services such as Lakera?
A guardrail service is usually a second call to a second vendor, made alongside the model call. In Vidai the guardrails are part of the call itself: pattern checks in the gateway and, on Enterprise, VidaiGuard's own small models beside it, all inside your network. There's no extra round trip, and one record holds both the call and the decision.
How does Vidai compare with other AI gateways, such as LiteLLM, Portkey or Kong?
Gateways like these are strong at connecting applications to many models. Vidai's focus is governance: regulations applied to every call, a signed record for auditors, cost per agent with budgets that act, and its own guardrail models, all inside your network. The full side-by-side is on vidai.uk/ai-gateway-comparison.
Is Vidai neutral between model providers?
Yes. We sell no models and no agents, so every provider is governed the same way, your own models included. Switching provider is a setting, not a rewrite: your apps keep the SDK they were written for.
Does Vidai check the answers that come back from a model?
Guardrails check everything sent to a model, every message and tool results included, before it leaves. The answer coming back isn't checked; it's recorded with the call, and where a prompt was pseudonymised, the original values are put back into the answer. Checking answers is where security tools add to Vidai.
Deployment & security
Where does Vidai run?
Inside your own network, your cloud account, your data centre, or fully air-gapped. It is not a hosted service your traffic passes through.
Does it send any data back to Vidai?
No. There is no telemetry, no usage callback and no outbound beacon: it does not phone home, and you can verify that on the wire. The one optional connection is for model prices: our public rate-card server, maintained on a best-effort basis. Vidai only fetches price lists from it, and you can host your own rate-card server (Enterprise) or enter prices by hand instead.
Can it run in an air-gapped environment?
Yes. No outbound connectivity is required to operate, so it can be deployed where there is no internet access at all. Prices come from a rate-card server you host inside the network (Enterprise), or from prices entered by hand.
Does it integrate with our SSO and identity model?
Yes. Single sign-on works with any OpenID Connect provider, such as Microsoft Entra ID, Okta, Google Workspace and Keycloak, and your groups map to teams and roles in the Control room. Single sign-on comes with Enterprise; multi-factor sign-in is in every edition.
How big is it to deploy?
The engine that sits in your AI traffic is about 25 MB; Vidai runs as containers with PostgreSQL, inside your network. The Control room runs alongside it. There is no large platform to stand up.
Commercial
How is Vidai priced?
Talk to us about Scale and Enterprise; Community is free. There are no charges per token, per seat, per call or per logged request, and no usage markup, so the bill doesn't grow just because your traffic did. The editions are compared at /editions.
Is Vidai open source?
Vidai, the control plane, is a commercial product. VidaiMock, the LLM mock server for CI, is open source.
Is there a free or community edition?
Yes. The Community edition is the full governance-first control plane, in your own network, free, no expiry. Sign in at portal.vidai.uk to generate your licence and download the latest release from github.com/vidaiUK/vidai-quickstart. Commercial use is by approval, granted case-by-case. Start at /community.
Do you offer a trial or proof of concept?
Yes. It starts with a 20-minute demo on infrastructure that looks like yours. Then, if it fits, we agree a two-week pilot on your own traffic: one app or team, your infrastructure, rules in observe mode first.
Working with us
Can you build a guardrail model for our own risks?
Yes. We build small, dense guardrail models from your own data, with the recipe behind our own family of six, two of them open weights on Hugging Face. Each is one portable ONNX file, fast enough to check every call, and the model, its training pipeline and its evaluation are yours. See vidai.uk/services.
Do you help us deploy it?
Yes. Our services team handles deployment, policy and cost setup, and can build custom guardrail models or migrate a legacy stack. Every engagement ends with something you own and run.
Where are you based?
Glasgow, UK. The team's production experience spans operators like Amazon, GE and Tata.
How do we start?
Request a demo, or start with a Sovereignty & Cost Audit to see where you stand. No procurement process to begin a conversation.
Product names are trademarks of their owners, named as examples of their category. What we say about other products comes from their own public documentation and announcements, as of November 2026.
How we start together
Still have a question.
- 01
Pick one app or team
Point it at the Control room. One configuration change; nothing else changes.
- 02
Watch for two weeks
Every call priced, checked and recorded. Rules start in observe mode.
- 03
See the evidence
What it cost, what it caught and what it would have stopped, from your own traffic.