Live · v0.0.1 preview

Stop your AI agent from making promises it can't keep.

TrustLoopGuard is a runtime guardrail plugin that intercepts agent output, applies your policies, and ships an immutable audit trail. Local-first. Sub-50ms. Three lines to install.

Decisions today

Intercepts

11ms

p95 latency

tlg-runtime · session conv_4821 · policies/refund-promise.yaml

Customerinbound

Agent draftmodel output

Delivered post-guard — pending

Audit log

Run a scenario to populate the audit trail.

A plugin, not a platform.

TrustLoopGuard runs inside your agent process. Policies are YAML in your repo. Decisions are local. The cloud backend is optional — only sync if you want shared policies and a team dashboard.

i.
Install
One package. Wraps your agent's response handler. No proxy, no sidecar.
ii.
Author policies
YAML files committed to your repo. Regex, literal, or LLM-graded matches.
iii.
Ship the audit trail
Every decision logged with input → matched policy → action. Replayable.

tlg-sdk · 0.0.1

# pip install trustloopguard
from trustloopguard import Guard
from my_agent import agent

guard = Guard.from_policies("./policies")

async def handle(message):
    draft = await agent.respond(message)
    verdict = await guard.check(draft, channel="chat")
    return verdict.delivered  # rewritten or original

// npm install @trustloopguard/sdk
import { Guard } from "@trustloopguard/sdk";
import { agent } from "./agent";

const guard = await Guard.fromPolicies("./policies");

export async function handle(message) {
  const draft = await agent.respond(message);
  const verdict = await guard.check(draft, { channel: "chat" });
  return verdict.delivered;
}

id: refund-promise
when:
  channel: [voice, chat]
match:
  any:
    - regex: "(?i)\b(refund|money[- ]back|guarantee)\b"
    - literal: "I promise"
action: rewrite
rewrite: "I'll connect you with a teammate who can review that for you."
severity: high