Guardrails AI on Future AGI
Guardrails & Safety
Schema-based input/output validation for LLM apps.
What you get
Everything traced, scored, and improvable
One install instruments Guardrails AI on OpenTelemetry. Future AGI then layers evaluators, optimisers, and simulations on top of the same trace tree — no second SDK, no double instrumentation.
Trace
Auto-instrumented spans
Every Guardrails AI call becomes a span — inputs, outputs, latency, tokens, cost, model name, tool args, retrieval results, and chain steps captured automatically.
Evaluate
70+ evaluators on every span
Attach Groundedness, Context Relevance, Prompt Injection, Toxicity, and 70+ more — purpose-built scorers powered by the Turing eval models, not generic LLM-as-judge.
Optimize
Closed-loop improvement
Pipe failed traces into agent-opt: GEPA, PromptWizard, ProTeGi, and Bayesian search rewrite your prompts with proof of measured gains.
Simulate
Adversarial scenarios at scale
Generate hundreds of personas and run them through your Guardrails AI agent before launch — text and voice, scripted or persona-driven.
Quickstart · <3 min
Instrument Guardrails AI in three steps
-
Step 1
Install the traceAI package
One package per language, ships from PyPI, npm, and Maven Central.
-
Step 2
Register the trace provider
Set
FI_API_KEYandFI_SECRET_KEY, then callGuardrailsInstrumentor().instrument(). -
Step 3
Run your existing Guardrails AI app
No code changes. Traces appear in the Future AGI dashboard within seconds.
Install
pip install traceAI-guardrailsInstrument
from fi_instrumentation import register
from fi_instrumentation.fi_types import ProjectType
from traceai_guardrails import GuardrailsInstrumentor
trace_provider = register(
project_type=ProjectType.OBSERVE,
project_name="GUARDRAILS_APP",
)
GuardrailsInstrumentor().instrument(tracer_provider=trace_provider)
# Your existing Guardrails AI code runs unchanged from here.
# Every call is now an OpenTelemetry span in Future AGI.Pick a recipe
What do you want to do with Guardrails AI?
Each recipe is a copy-paste page with the exact code, the gotchas, and a working example you can clone.
Recipes for Guardrails AI