A gate for support agents before they go live

I almost let a support bot approve a refund it shouldn't have. Agent Gate is the gate I wished existed: run golden tasks before deploy, flip a kill switch, keep a ledger managers can read.

gate: active · eval pack: support-v1 · ledger: append-only

01 — Demo

What you'll see

Run eval on the staging support agent — it passes. Run the same pack on a broken prompt — score drops. Hit Deploy on the failed one — blocked. That's the whole product.

Eval status
Support Copilot (staging)
eval 0% · draft
Support Copilot (broken prompt)
eval 0% · draft
Threshold
80%
Ledger
eval_completedstaging agent — 100%8s ago
deploy_blockedbroken prompt — score below 80%5s ago

Open the live demo →

02 — Audience

Who it's for

Teams already running support agents on Zapier, n8n, Lindy, or custom code. Not another chatbot builder — a thin layer outside your existing stack.

Webhook ingest on day one. No full rewrite. First wedge is customer support: refunds, escalations, PII.

Zapiern8nLindycustom

03 — Partners

Design partners

I'm looking for 3 people who already run (or are shipping) a support agent. I'll wire Agent Gate to your workflow in a 30-minute call — eval pack, kill switch, webhook ledger. Free for 2 weeks while you give blunt feedback. After that, $99/mo or walk. No hard sell.

Prefer email? trefongwork@gmail.com