Skip to content

docs(guardrails): add the LLM Shield Proxy guardrail page - #2070

Merged
yucheng-berri merged 2 commits into
BerriAI:mainfrom
ninadphalak:docs/llm-shield-proxy-guardrail
Oct 5, 2026
Merged

yucheng-berri merged 2 commits into
BerriAI:mainfrom
ninadphalak:docs/llm-shield-proxy-guardrail

Conversation

@ninadphalak

Copy link
Copy Markdown
Contributor

Adds the guardrail page for LLM Shield Proxy (guardrail: llm_shield_proxy), which accompanies BerriAI/litellm#42645, and lists it in the guardrail providers sidebar.

The page covers running the Shield, the config entry with both pre_call and post_call on one guardrail, a first request in plain and streaming form, the fail-closed error, which request fields are redacted into the restorable and non-restorable vaults, and the two parameters. The example output and the error body are taken from a run of #42645 against the published Shield image.

npm run lint:docs and npm run lint:writing pass, and npm run build succeeds with the new page.

@vercel

vercel Bot commented Oct 4, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
litellm Ready Ready Preview Oct 5, 2026 5:56pm UTC

Request Review

…tation

- cover /v1/completions (request walk and reply restore)
- document the newer redaction surfaces: extra_body, Anthropic document
  parts, Responses tool outputs / custom tool input / code-interpreter
  code / typed prompt variables, system+developer items in Responses input
- new section: model-level SDK usage (deployment-hook redact+restore,
  response-cache bypass) and the fail-closed refusal of SDK streams
- note the proxy-path caching and guardrail-telemetry behavior
@yucheng-berri

Copy link
Copy Markdown
Contributor

Synced this page with the guardrail's current implementation (head db4191254 of BerriAI/litellm#42645):

  • /v1/completions added to the covered endpoints (request walk + reply restore)
  • Newer redaction surfaces documented: extra_body, Anthropic document parts, Responses tool outputs / custom tool-call input / code-interpreter code / typed prompt variables, and system+developer items in Responses input
  • New "Using it with the LiteLLM SDK" section: model-level usage via guardrails=[...], the response-cache bypass, and the fail-closed refusal of SDK stream=True
  • Short "Caching and telemetry" note for the proxy path (cache stores the redacted reply; guardrail telemetry records stand-ins, not restored values)

All statements verified live against a running proxy + LLM Shield container during the audit of the code PR.

@yucheng-berri
yucheng-berri enabled auto-merge (squash) October 5, 2026 17:40
@yucheng-berri
yucheng-berri merged commit a0035d8 into BerriAI:main Oct 5, 2026
2 checks passed
ninadphalak added a commit to ninadphalak/LLM-Shield-Proxy that referenced this pull request Oct 6, 2026
…#188)

* docs(litellm): point to LiteLLM's built-in llm_shield_proxy guardrail

LiteLLM merged the guardrail (BerriAI/litellm#42645) and its docs page
(BerriAI/litellm-docs#2070) on 2026-10-05. Our LiteLLM page still said
the integration had no dashboard card and no LiteLLM docs page.

* docs(litellm): name the merge commit instead of a predicted release

This branch was successfully deployed

1 active deployment
Preview — 18431499 Deployed Oct 5, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants