Use cases

Call OpenAI, Anthropic, Azure, Bedrock, Vertex, and Google through one OpenAI-compatible endpoint, with failover and no lock-in.

Cursor, Claude Code, and Codex behind one governed endpoint, with a budget on every key and cost attributed per developer.

Attribute every call to a person, team, project, and cost center, gross and net of what aigw saved you.

Request logs plus OpenTelemetry traces and metrics: the model, tokens, cost, latency, and policy outcome of every call.

Detect and act on PII, secrets, and national IDs in prompts and responses, before they leave for the model provider.

Detect and block injection and jailbreak attempts, and scan model, tool, and agent output for exfiltration.

Give every team one sanctioned endpoint, so every call is authenticated, governed, and logged instead of scattered across personal keys.

Argument-level guardrails, human approvals, and secret scanning on every tool call, not just chat completions.

Control which agents may delegate to which, require approvals, rate-limit hops, and scan every response.

Roles and per-key model allow-lists, mapped to your SSO groups, so everyone gets least-privilege access.

Platform teams set the models, budgets, and policies once; teams onboard themselves without filing a ticket.

An EU-hosted control and data plane, per-model region routing, and credentials that never leave the gateway.

A tamper-evident audit trail and a per-call governance record for the EU AI Act, GDPR, NIS2, and ISO 27001 reviews.