Files
netbird/e2e/agentnetwork
mlsmaycon b4a9e1c5c8 e2e: per-group model allowlist across all provider surfaces
Add TestGuardrailPerGroupAllowlist_AllProviders: for each always-on provider
surface — OpenAI (WireChat), Anthropic (WireMessages), Vertex (URL path) and
Bedrock (URL path, normalized) — two policies target the provider, granting the
client's group only the selected model and a second group only the other model.
The client gets the selected model served (200) and the other group's model
denied (403 llm_policy.model_blocked), proving the per-policy/group decision
holds across body- and path-routed providers (the cross-group no-leak property).
2026-07-26 18:15:12 +00:00
..