Contact us
Tell us what AI workflow you need to protect, which regional boundary applies, and what evidence your team needs before it can trust a gateway. We can discuss a bounded pilot while self-service registration remains closed.
Public documentation, measured evidence, the constitution, threat radar, and the appropriate playground remain open to read. This page starts a conversation only: it does not create an account, API key, checkout, payment, deployment authorization, or launch commitment.
Who a pilot is for
A useful pilot begins with a concrete risk and a workflow that can be tested without exposing production credentials or customer data.
- Product and platform teams putting chat, RAG, model routing, or agent tools behind an API boundary and needing visible policy decisions.
- Security teams evaluating prompt-injection, jailbreak, data-exposure, tool-use, auditability, and fail-closed behavior with reproducible evidence.
- Regulated, sovereign, or infrastructure operators that need RU and Global identities, providers, callbacks, credentials, and data paths kept strictly separate.
What to include in the first message
High-level context is enough for the first pass. These details let us decide whether a safe, bounded evaluation is currently possible.
- Required contour and jurisdiction: RU, Global, or an isolated self-hosted environment, plus any hard data-residency constraint.
- The workflow: prompt or response scanning, chat gateway, RAG, an agent or MCP tool chain, and the main failure you want to prevent.
- Expected deployment model, identity provider requirements, mail or webhook dependencies, and whether the evaluation must avoid external model calls.
- A measurable outcome: block or observe a named attack class, produce reviewable evidence, measure false positives, or validate an integration boundary.
What happens next
We first check product fit, regional dependencies, and whether the requested journey can be tested honestly. If a required identity provider, staging target, mailbox, rollback path, or live canary is unavailable, we say so and keep that capability closed. Any pilot scope, data handling, and success criteria are agreed explicitly before testing; public interest is never converted into production access or a commercial commitment by default.