[bot] Instrument Guardrails AI (30,625 weekly downloads) #807

Summary

guardrails-ai (import name guardrails, by Guardrails AI Inc.) is a framework for adding structural, type, and quality validation directly around LLM calls. Its primary execution surfaces are Guard.for_pydantic(output_class=...) / Guard().use(<validators>) to construct a guard, and then the callable form guard(model=..., messages=[...]), which invokes the LLM directly and returns a response with .validated_output — i.e. the Guard itself owns the LLM call, not just post-hoc validation of already-generated text. guard.parse(...) is the sibling entrypoint for validating a given string against attached validators.

This is a distinct project/package from NVIDIA NeMo Guardrails (nemoguardrails, tracked separately in #768) — different maintainer, different PyPI package, different API shape (Guard/validators around a direct LLM call vs. NeMo's LLMRails rails pipeline). This repository has zero instrumentation for Guardrails AI anywhere: no py/src/braintrust/integrations/guardrails* directory, no wrapper, no nox session, no cassette directory, and no entry in py/pyproject.toml's [tool.braintrust.matrix].

What's missing

• guard(model=..., messages=[...]) (and async variant) — the primary "guarded LLM call" entrypoint; since Guardrails AI drives the LLM call itself, an unwrapped call here means the underlying model call, validated output, and any validator retries are entirely invisible to Braintrust tracing.
• guard.parse(...) / guard.parse_async(...) — validates a given string against attached validators, returning a result with .validation_passed.
• Guard.for_pydantic(output_class=...) / Guard().use(<validators>) — guard construction with attached validators (useful for capturing validator configuration as metadata on the guarded-call span).

Weekly downloads

Weekly downloads: 30,625 (as of 2026-09-28; https://pypistats.org/api/packages/guardrails-ai/recent)

Braintrust docs status

not_found. Checked https://www.braintrust.dev/docs/guides/tracing and https://www.braintrust.dev/docs/guides/tracing/integrations (both checked 2026-09-28) — neither the tracing-integrations list ("OpenTelemetry, Temporal, Vercel AI SDK, OpenRouter SDK, LangChain, LangChain4j, LangSmith, LlamaIndex, Apollo GraphQL, Cloudflare Workers AI, DSPy, Google Discovery Engine, Instructor, LiteLLM, Ruby LLM, Spring AI, Traceloop, TrueFoundry") nor the agent-framework list mentions Guardrails AI or Guardrails.

Upstream sources

• PyPI: https://pypi.org/project/guardrails-ai/
• GitHub: https://github.com/guardrails-ai/guardrails
• Docs (quickstart): https://www.guardrailsai.com/docs/getting_started/quickstart

Braintrust docs sources checked

• https://www.braintrust.dev/docs/guides/tracing (checked 2026-09-28)
• https://www.braintrust.dev/docs/guides/tracing/integrations (checked 2026-09-28)

Local repo files inspected

• py/src/braintrust/integrations/ — no guardrails/ directory (only unrelated py/src/braintrust/integrations/__init__.py and the separately-tracked nemoguardrails gap in [bot] Instrument NVIDIA NeMo Guardrails (149,615 weekly downloads) #768)
• py/src/braintrust/wrappers/ — no Guardrails AI wrapper
• py/pyproject.toml [tool.braintrust.matrix] — no guardrails-ai entry
• py/noxfile.py — no test_guardrails session
• Repo-wide case-insensitive grep for guardrails.ai/guardrails_ai under py/src/braintrust/**/*.py — zero matches