Integration
Fact-checking & guardrails for Claude
Claude models are the workhorses of a lot of serious LLM deployments – and none of Anthropic’s safety work changes what happens at your boundary: your PII still enters the prompt unless something screens it, a summarized figure can still drift from its source, and your compliance duties are still yours. The proxy adds that boundary without asking you to change how you build.
Two routes, depending on what you need. The native Anthropic adapter forwards to the Anthropic API directly and covers the full text pipeline – PII screen and restore, grounding and fact-checking, format enforcement, content safety, injection defense. For tool-using agents there is a practical wrinkle: the proxy’s tool-call inspection works on OpenAI-compatible tool structures, so agents on Claude get the full agent stack – tool policy, permission tiers, action grounding, budgets – by routing through OpenRouter, which serves Claude models over that surface. Same models, same proxy, one line different.
client = OpenAI(
base_url="https://api.inguardout.com/v1",
api_key="gr-…",
)
client.chat.completions.create(
model="claude-sonnet-5", # served via your configured upstream
messages=[…],
)1. Pick the route
Text workloads: point a proxy key at the Anthropic API with your Anthropic credentials. Tool-using agents: point it at OpenRouter and select Claude models there – full tool-call inspection included.
2. Keep your client
Your app talks to the proxy over the OpenAI-compatible surface either way – one base URL and key swap, no Anthropic-specific client code needed.
3. Turn on the checks
PII screening, grounding, format, and safety run on every Claude response; agent checks engage on tool calls when routed via OpenRouter.
4. Watch the tiers
Route Opus-class work and Haiku-class work through different keys with different check profiles – stricter grounding where the cheaper tier answers.
Which Claude models work?
Whatever your configured upstream serves – Opus, Sonnet, and Haiku tiers via the Anthropic API, and the same families via OpenRouter. The proxy is model-agnostic; pricing metadata drives per-request cost tracking.
Doesn’t Claude already have strong safety built in?
Model-side safety and boundary enforcement answer different questions. Anthropic’s alignment work governs what the model will say; it cannot screen your customers’ PII out of prompts, verify figures against your source documents, enforce your tool policies, or produce your audit trail. Those live at your boundary.
Why do agent checks need the OpenRouter route?
The proxy inspects tool calls in OpenAI-compatible structures. The native Anthropic adapter is text-only today, so tool_calls pass through it uninspected; OpenRouter serves Claude over the OpenAI-compatible surface, which gives the agent stages everything they need.
Does PII screening interfere with Claude’s answers?
Placeholders are stable per value, so the model reasons over consistent tokens and the restore step returns a natural answer. Grounding compares against the screened source, so placeholders are never mistaken for changed values.
Put guardrails in front of Claude
Route your Anthropic calls through the proxy to screen PII, catch hallucinations, and enforce output. We are running a limited demo - sign up and we will get you in as soon as we can.