· Live walkthrough ·
Copy-paste ready demo – Stripe → Lemon Squeezy fallback in < 60 seconds
A 1-minute demo of toolrate.guard() in action — the agent consults ToolRate, sees a 62/100 reliability score, and automatically falls back to a 96/100 alternative. No retries. No wasted tokens. No duplicate charges.
1 minute · turn sound on for the live commentary
62 / 100
Stripe reliability
96 / 100
Fallback chosen
~40K
Tokens saved
Copy-paste ready · run locally in one command
The exact script from the video. Pure Python, no dependencies, no API keys.
python demo_toolrate.py
#!/usr/bin/env python3
"""
ToolRate Loom demo — single-file simulation.
Run:
python demo_toolrate.py
Tip: maximize your terminal, dark background, ~14pt font.
Tune SPEED to stretch/shrink the runtime (1.0 ≈ 70 seconds).
"""
import sys
import time
SPEED = 1.0 # global pause multiplier — bump to 1.2 if you talk slowly
# ---------- ANSI ----------
RESET = "\033[0m"
BOLD = "\033[1m"
DIM = "\033[2m"
RED = "\033[91m"
GREEN = "\033[92m"
YELLOW = "\033[93m"
BLUE = "\033[94m"
MAGENTA = "\033[95m"
CYAN = "\033[96m"
GRAY = "\033[90m"
WHITE = "\033[97m"
BG_RED = "\033[101m"
BG_GREEN = "\033[102m"
WIDTH = 72
def pause(seconds):
time.sleep(seconds * SPEED)
def slow(text, delay=0.014, end="\n"):
for ch in text:
sys.stdout.write(ch)
sys.stdout.flush()
time.sleep(delay)
sys.stdout.write(end)
sys.stdout.flush()
def rule(char="─", color=GRAY):
print(f"{color}{char * WIDTH}{RESET}")
def banner(title, color=CYAN):
rule("━", color)
print(f"{color}{BOLD} {title}{RESET}")
rule("━", color)
def think(text, delay=0.9):
print(f" {DIM}{GRAY}↳ {text}{RESET}")
pause(delay)
def step(symbol, text, color=WHITE, delay=0.5):
print(f" {color}{symbol}{RESET} {text}")
pause(delay)
# ---------- DEMO ----------
def main():
# clear screen
sys.stdout.write("\033[2J\033[H")
sys.stdout.flush()
# ── session header ─────────────────────────────────────────────
print()
print(f"{BOLD}{CYAN} ◆ Claude Code{RESET} {GRAY}— agent session{RESET}")
print(f" {GRAY}model: claude-opus-4-6 · tools: stripe, lemonsqueezy, toolrate{RESET}")
print()
rule()
pause(0.8)
# ── user task ──────────────────────────────────────────────────
print()
print(f" {BOLD}user ▸{RESET}")
slow(" Charge the customer $49 for a premium subscription using Stripe.")
print()
pause(1.6)
# ── agent reasoning ────────────────────────────────────────────
print(f" {BOLD}{MAGENTA}assistant ▸{RESET}")
pause(0.4)
think("Understanding task: process a $49 premium subscription charge…")
think("Identifying tool: Stripe Payments API → /v1/charges")
think("Policy: always guard external API calls with ToolRate before invoking.")
print()
pause(0.6)
# ── toolrate.guard call ────────────────────────────────────────
step("⚡", f"{BOLD}toolrate.guard(\"stripe.charges.create\", fallbacks=\"auto\"){RESET}", CYAN)
pause(0.6)
print(f" {GRAY}POST https://toolrate.ai/v1/assess{RESET}")
pause(0.5)
print(f" {GRAY}↻ analyzing 1,284 recent reports…{RESET}")
pause(1.4)
print()
# ── toolrate response card ─────────────────────────────────────
banner(" TOOLRATE ▸ reliability oracle", CYAN)
print()
print(f" {BOLD}tool{RESET} stripe.charges.create")
print(f" {BOLD}reliability{RESET} {BG_RED}{WHITE}{BOLD} 62 / 100 {RESET} "
f"{RED}↓ below safe threshold (80){RESET}")
print(f" {BOLD}risk_level{RESET} {RED}{BOLD}MEDIUM-HIGH{RESET}")
print(f" {BOLD}sample_size{RESET} 1,284 reports · last 7 days")
print()
pause(2.0)
print(f" {BOLD}{YELLOW}pitfalls{RESET}")
print(f" {YELLOW}•{RESET} Missing {BOLD}idempotency_key{RESET} → duplicate charges on retry")
pause(0.5)
print(f" {YELLOW}•{RESET} Test-mode vs live-mode key mixup ({BOLD}19%{RESET} of failures)")
pause(0.5)
print(f" {YELLOW}•{RESET} 3DS challenge not handled → silent decline in EU")
print()
pause(1.4)
print(f" {BOLD}{BLUE}mitigations{RESET}")
print(f" {BLUE}→{RESET} pass Idempotency-Key header (uuid4)")
print(f" {BLUE}→{RESET} verify sk_live_* prefix before request")
print(f" {BLUE}→{RESET} handle requires_action / 3DS flow")
print()
pause(1.4)
print(f" {BOLD}{GREEN}top_alternative{RESET}")
print(f" lemonsqueezy.checkout.create "
f"{BG_GREEN}{WHITE}{BOLD} 96 / 100 {RESET} {GREEN}✓ recommended{RESET}")
print(f" {GRAY}same-intent · handles tax · idempotent by design{RESET}")
print()
rule()
pause(2.6)
# ── decision ───────────────────────────────────────────────────
print()
print(f" {BOLD}{MAGENTA}assistant ▸{RESET}")
think("Reliability 62 < threshold 80 — Stripe path is unsafe right now.")
think("fallbacks=auto enabled → switching to top_alternative.")
print()
step("↻", f"{YELLOW}Switching:{RESET} stripe → {BOLD}{GREEN}lemonsqueezy{RESET}", YELLOW)
pause(1.0)
print()
# ── execution ──────────────────────────────────────────────────
step("⚡", f"{BOLD}lemonsqueezy.checkout.create(amount=4900, currency=\"usd\"){RESET}", CYAN)
pause(0.6)
print(f" {GRAY}→ creating checkout…{RESET}")
pause(0.8)
print(f" {GRAY}→ charging card ending 4242…{RESET}")
pause(1.0)
print()
# ── success ────────────────────────────────────────────────────
print(f" {GREEN}{BOLD}✅ Payment successful{RESET} — "
f"{BOLD}$49.00{RESET} charged via {GREEN}{BOLD}Lemon Squeezy{RESET}")
print(f" {GRAY}order_id: ls_ord_8f3a21bc · status: paid{RESET}")
print()
pause(1.0)
print(f" {GREEN}{BOLD}✓ Task completed successfully{RESET}")
print()
print(f" {DIM}{GRAY}─ avoided ~40,000 tokens of debug + retry loops{RESET}")
print(f" {DIM}{GRAY}─ avoided a potential duplicate charge{RESET}")
print(f" {DIM}{GRAY}─ guarded by toolrate.ai before the first byte was sent{RESET}")
print()
pause(2.2)
# ── CTA ────────────────────────────────────────────────────────
rule("━", CYAN)
print()
print(f" {BOLD}{CYAN} Try ToolRate{RESET} {WHITE}▸{RESET} "
f"{BOLD}https://toolrate.ai{RESET}")
print(f" {GRAY} 600+ tools rated · free tier · pip install toolrate{RESET}")
print()
rule("━", CYAN)
print()
pause(2.0)
if __name__ == "__main__":
try:
main()
except KeyboardInterrupt:
print(f"\n{GRAY}— demo aborted —{RESET}")
Based on 1,284 real agent reports · Hosted in Germany · pip install toolrate
Make your agents this reliable — today.
Free tier gives you 100 assessments per day. No credit card. Drop one
toolrate.guard()
call in front of any tool and watch failure rates collapse.