Layered defenses — rate limits, CSP, and circuit breakers — that protect the platform from abuse, control costs, and keep service reliable for all customers.
Running an AI assistant at scale means you're operating a service that could, in theory, be misused — flooded with automated requests, probed for weaknesses, or manipulated through malicious inputs. This feature covers the layered defenses that keep your platform secure, your costs predictable, and your customers' experiences uninterrupted: per-customer usage limits, content security policies, and automatic circuit breakers that protect your system when something upstream goes wrong.
Security and reliability are table stakes for enterprise sales. Customers want to know their data is protected, their usage won't be affected by someone else's abuse, and that the platform won't rack up unexpected costs on their behalf.
Each API key has a quota — a maximum number of requests allowed within a given time window. Every request checks this quota in real time:
- Under the limit: request goes through normally
- Approaching the limit: the response includes headers showing how much quota remains, so the calling system can slow down gracefully
- Over the limit: the request is rejected with a clear error, not silently dropped
Quotas can be set per customer, per integration, or globally. Legitimate customers who hit their limit by accident can request an increase through a self-service flow.
Every page served by the platform includes instructions to browsers about what scripts and resources are allowed to load. If a browser detects a violation — say, an injected script trying to load from an unauthorized source — it reports that violation back to the platform. These reports are aggregated and can trigger alerts if a pattern of violations is detected.
If an upstream service (like an AI provider) starts returning errors or responding slowly, the circuit breaker automatically reduces traffic to that provider and routes requests to a backup. Once the provider recovers, traffic is gradually restored. This is invisible to end users — they experience a brief slowdown at worst, not a hard failure.
The team is building:
Once live, you'll have clear controls and visibility over platform usage, with automatic protection against both external abuse and upstream failures.