Per-project security guardrails and expanded multi-modal coverage

Multi-modal support now extends to audio and video models

New

Agents can now route requests to models built for audio or video input through the same Gateway endpoint they already use for text and image models. Read more in the Gateway docs.

Prompt Injection Protection and DLP guardrails now configurable per project

Security

Each project can now set its own Prompt Injection Protection thresholds and DLP rules, overriding the organization default where it needs to. A project can only tighten the organization's policy, never loosen it, so the org-wide setting stays the floor for every project under it. Read more in the Gateway security docs.