From 0bd6e5bff8f577d46cd01a041e7b69d203c6d5af Mon Sep 17 00:00:00 2001 From: albertbausili Date: Wed, 2 Sep 2026 12:26:16 +0200 Subject: [PATCH] docs(guardrails): link the NeuralTrust setup guide from the hook README Points at the TrustGuard integration page for the setup walkthrough, the verdict mapping, and the streaming caveat, so the README can stay a reference rather than repeat it. Adds a References section matching the one on the IBM Guardrails hook, and uses the guardrails quick_start URL because the bare /docs/proxy/guardrails path 404s. --- .../guardrails/guardrail_hooks/neuraltrust/README.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/litellm/proxy/guardrails/guardrail_hooks/neuraltrust/README.md b/litellm/proxy/guardrails/guardrail_hooks/neuraltrust/README.md index 744a8f819c0..1958e6bb290 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/neuraltrust/README.md +++ b/litellm/proxy/guardrails/guardrail_hooks/neuraltrust/README.md @@ -2,6 +2,9 @@ Native LiteLLM guardrail. Sends chat input and output to TrustGuard `POST /v1/evaluate`. +Setup guide, verdict mapping, and the streaming caveat: +[docs.neuraltrust.ai/trustguard/integrations/litellm](https://docs.neuraltrust.ai/trustguard/integrations/litellm). + ## Config ```yaml @@ -43,3 +46,10 @@ HTTP 503 entitlements, 401/403, other 4xx/5xx, and unusable TrustGuard verdicts ## Streaming LiteLLM streaming guardrails default to `block_only`. `block` still fires on streamed calls. `transform` rewrites are not applied to the streamed tokens; use non-streaming requests when DLP redaction must reach the client. + +## References + +- [NeuralTrust TrustGuard on LiteLLM](https://docs.neuraltrust.ai/trustguard/integrations/litellm) +- [TrustGuard Evaluate API](https://docs.neuraltrust.ai/trustguard/api/evaluate) +- [TrustGuard collectors](https://docs.neuraltrust.ai/trustguard/concepts/collectors) +- [LiteLLM Guardrails Documentation](https://docs.litellm.ai/docs/proxy/guardrails/quick_start)