docs: LLM-as-a-Judge guardrail - #17
Merged
mubashir1osmani merged 8 commits intoMay 21, 2026
Merged
Conversation
New guardrail type that uses an LLM to score responses against weighted criteria. Includes UI walkthrough, YAML config examples, blocked/passed response examples, and configuration reference.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…d guardrail views
* docs(blog): make /blog responsive - Add mobile styles to swizzled BlogListPage (hero, marquee, posts, pagination). - Fix horizontal overflow caused by the marquee's white-space: nowrap propagating width up the flex chain. Break it with min-width: 0 on .page and #__docusaurus > *, plus defensive overflow-x: clip. - Respect prefers-reduced-motion (stop marquee animation). * Remove global overflow prevention styles Removed global styles to prevent horizontal overflow on mobile.
…ides (#31) Backfills the `BerriAI/litellm-docs` site with the changes that originally shipped under `docs/my-website/` in BerriAI/litellm#25939 / #26325. After the docs source was migrated to this repo, those edits could no longer be carried in the code PR and were dropped on rebase. - proxy/token_auth.md: expand "Matching behavior" with AND semantics, the new optional `scope` selector, list/string forms, shell-style wildcards (`*`, `?`, case-sensitive), and the scope-only space-split rule (iss/aud/client_id are never split on spaces). Adds a worked example combining `scope` with a wildcard `client_id`. - proxy/oauth2.md: cross-reference the new wildcard/scope behavior and point readers to token_auth.md for full details. Code change is in BerriAI/litellm#26325 (litellm_internal_staging).
…) (#184) * docs(mcp,a2a): code-verified auth reference fixes + overview page (#156) * docs(mcp): complete auth_type table, OAuth config reference, RBAC intersection model, hub-vs-public-internet distinction * docs(a2a): document x-litellm-api-key, trace-id enforcement, sub-agent propagation, agent access groups and full intersection model * docs(bedrock_agentcore): add LiteLLM A2A Gateway section — fixes broken anchor from a2a.md, documents dual JWT/SigV4 auth modes and full credential chain * docs: add AuthN/AuthZ overview page side-by-siding MCP and A2A gateways * docs(fixup): corrections from code-review pass — verified against current LiteLLM source * make changes --------- Co-authored-by: michelligabriele <gabriele.michelli@icloud.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds documentation for the new
llm_as_a_judgeguardrail type.What's in this PR:
proxy/guardrails/llm_as_a_judge.mdHow it works: After each LLM response, a judge model scores it against weighted criteria (e.g. Honesty 60%, Helpfulness 40%). If the overall score < threshold, the request is blocked with HTTP 422 and the full verdicts returned. If
on_failure: log, it passes through for monitoring.Tested: blocked call scored 6.0/100, passing call scored 95/100.