Unlike general-purpose API gateways, Nexus is purpose-built for organizations running AI under regulatory oversight—like enterprises, critical infrastructure, and government programs that need to follow strict rules about AI and data security.

Nexus creates a permanent, tamper-proof record of every request. If something goes wrong with a security rule, the system blocks the request instead of letting it through. Labels on your data make sure sensitive information stays where it’s supposed to. These built-in controls give organizations the compliance they need, without having to retrofit security onto tools that weren’t designed for it.

Nexus is built for enterprises, critical infrastructure, government programs and you where policy enforcement, provable audit trails, and end-to-end tracing of AI activity are hard requirements.

• AI gateway: Reach OpenAI and Anthropic through one OpenAI-compatible API. Chat, Responses, embeddings, images, files, and vector stores.
• Native Anthropic support: Anthropic SDKs and Claude Code work by setting ANTHROPIC_BASE_URL. No translation shims in your code.
• Routing and fallbacks: Pick models with a provider/model grammar; requests fall back across alternate providers and models with retries and optional response caching.
• Bring your own keys: Tenants register their own provider keys, encrypted at rest. Operators can add managed credentials with wallet-based billing.
• Cost tracking: Every request is priced per provider, model, service tier, and context band from a live model registry.
• Observability and agent tracing: A complete record of what your AI systems did. Inspect requests, sessions, and agent traces (OTLP) in the console, or export to your own collector and webhooks — the evidence you need for incident reviews, security investigations, and regulator inquiries.
• Policy engine: A dedicated policy decision point evaluates every request before it reaches a provider. Decisions carry machine-enforced obligations — field redaction, provider routing constraints, additional rate limits, audit tags, or an outright block — and services fail closed: an obligation that cannot be applied becomes a deny. Every evaluation emits its own audit event recording which policy decided, what it decided, and why.
• Classification enforcement: Every request carries a nexus-classification label validated against a per-scope vocabulary — workspaces seed the standard label set, and organizations and projects can extend it and tighten their own default and ceiling. Labels above the governing scope’s ceiling are denied before routing, with redaction obligations applied at the gateway — enforcement in the request path for environments that handle regulated or classified data.
• Tamper-evident audit: Per-tenant Merkle chains sealed with signed batch roots, WORM archival, and offline-verifiable inclusion proofs — an audit trail that meets the evidentiary bar of regulated and government deployments, with mechanisms mapped to NIST SP 800-53 audit controls in Compliance Map.
• Multi-tenant identity: Workspaces, organizations, and projects with RBAC, invitations, and OIDC sign-in.