A BAA is just the start of what it takes to safely run the AI your team is already using.
BAA coverage, audit logging, and access control
Cost and usage controls
Coming soon
De-identification and safety guardrails, defined in your code
View docs
Building for the AI-native era
AI is raising the bar on what attackers can do and what regulated buyers expect from their infrastructure. Some of LLM Gateway's highest-usage accounts already run Claude Code and other coding agents against regulated data, often without governance built specifically for that pattern. Here's what's coming soon to close that gap.
Budget alerts & usage reporting
Get alerted before a key hits its spend limit, with usage broken down by key, model, and scope.
Secure cloud-hosted agents
Run agents in a sandboxed environment with no open network access, routing every request through the Gateway.
Data residency
Keep all gateway infrastructure within a required region. Keys are automatically restricted to in-region models.
Product roadmap
Get through health system security reviews faster
Security review evidence is available for all model usage instantly, so you can turn a conversation that used to kill deals into a non-issue.
Protect PHI from model training
Training on PHI is disabled at the infrastructure layer; that protection holds regardless of any provider's own data-use policy or default settings.
Separate keys for internal tools and production AI
Give internal automations and agent harnesses their own keys so usage is attributable, with an org-wide spend limit that stops runaway loops before they become an incident.
Stay online when a model provider goes down
When a model is served by more than one provider, the gateway fails over automatically; you can also name backup models in the request and the gateway tries each in order.