$ the-wire · showcase
Guardrails streams tool calls, adds a language-detector rail
By RepoJournal · Filed · About Agent Security · Composed from the cited sources · methodology
NeMo Guardrails now accepts tool-calling requests in streaming mode and ships a standalone language-detector input rail, the two changes operators can wire into production today.
feat(library): add language detector input rail NVIDIA-NeMo/Guardrails
Guardrails' /v1/chat/completions endpoint now accepts tools with stream: true, generating and validating the full response first before emitting approved text and tool arguments as successive SSE deltas, so your tool rails and HTTP errors stay intact and malformed tool arguments get rejected. The same PR exposes the fast-langdetect logic already behind the content safety rail as a general-purpo...
fix(server): support streaming tool-call requests NVIDIA-NeMo/Guardrails
Operators running tool-calling agents behind Guardrails no longer have to choose between streaming and tool use on the chat completions endpoint; validation happens upstream of the stream, so a rejected response surfaces as an HTTP error rather than a half-delivered delta. This is the fix for the long-standing gap tracked as #2352.
feat(library): add language detector input rail (#2470) NVIDIA-NeMo/Guardrails
A new input rail filters inbound messages against a language allowlist or blocklist, driven by the same fast-langdetect code the content safety rail uses. Drop it in front of your pipeline if you need to refuse traffic outside your supported languages rather than answer it.
chore: prepare release 0.6.10 (#531) snyk/agent-scan
snyk/agent-scan moved to 0.6.10; the release prep carries no operator-facing detail in the commit, so pin the new version only if you track releases automatically.
