Skip to main content
Live · real AI, right now Flagship agent

The support agent that answers from your docs — and knows when to call a human.

Run a real ticket through it below — your words, your knowledge base. Watch it read the ticket, search your docs, and decide: answer instantly, or escalate with full context. It grounds every reply in what you gave it and refuses to invent. That guardrail is the whole point.

🤖 30%→67% auto-resolved as it learns 💸 ~$23k/yr saved 🌙 24/7 first response

Own the agent (an n8n workflow) and run it yourself, or have us build it on your stack. 30-day money-back guarantee.

Run the agent · live

Your ticket. Your docs. Its decision.

Pick a scenario or type your own. Paste a few lines of your real help docs if you want — or just name your business and leave the docs blank: the agent will draft sample docs for it, then answer from them. (Your live build grounds on your real docs.)

In plain English: a “grounded” answer is one the agent gives only using facts from your documents — it never makes things up. If the answer isn’t in your docs, it sends the ticket to a human instead of guessing.
Time of day

Runs on fast edge AI for the demo. The build you commission runs Claude or GPT with real RAG over your full docs.

FIG · customer support agent 🛡️ answers only from your docs · sends the rest to a human

🤖 Identifies itself as AI to customers · EU AI Act (Art. 50) ready

🌐 One agent, every channel — 💬 chat · ✉️ email · 📱 WhatsApp · 📞 voice · 🌍 replies in the customer's language

A new support ticket arrives by email, chat, or form and enters the workflow. WEBHOOKNew ticket A question comes in An AI step reads the ticket to understand intent and detects tone — frustration or sensitivity routes straight to a human. AI AGENTUnderstand Read intent & tone Searches YOUR knowledge base for a grounded answer — it never makes one up. VECTORSearch docs Ground in your KB The governance gate: resolve only with a confident, grounded, non-sensitive answer — otherwise escalate. IF · GATEResolve? Solve or escalate Sends the grounded answer back instantly — or hands a routed escalation to a human with full context. HTTPReply / hand off Answer or route Logs the outcome so deflection is measured and the agent keeps improving over time. POSTGRESLog & learn Track & improve

Agent activity

Press “Run the agent” — its reasoning streams here, step by step.

Everything inside this agent

A complete product — not a chatbot

Six things this agent does for you — in plain English. You just watched every one of them run live above. No jargon, no surprises.

🧠

It knows your business

Answers from your real help docs — and never makes things up.

  • · Uses your own help articles, policies and past answers
  • · No docs yet? It writes a starter set from a line about your business
  • · Shows the exact line each answer came from
  • · If the answer isn't there, it says so instead of guessing
🔧

It does things, not just chats

Handles the routine tasks that fill up your inbox.

  • · Looks up a customer's order, account or booking instantly
  • · Books, cancels or resets — the everyday requests
  • · Updates your helpdesk / CRM (Zendesk, HubSpot, Salesforce…)
  • · Anything risky waits for a person to approve it first
  • · Built so it can never do the same thing twice by accident
🛡️

Safe in front of customers

The guardrails that stop an AI embarrassing your brand.

  • · Hides card numbers & personal details before the AI sees them
  • · Ignores anyone trying to trick it into breaking its rules
  • · Checks who it's talking to before sharing private account info
  • · Keeps a full record of every decision (great for audits)
  • · Tells people they're talking to an AI (now required by law)
🏭

Works for any business

Ships with Governed Playbooks — your industry's rules, out of the box.

  • · One agent, ready for 12 industries
  • · Knows the rules for healthcare, finance, law, schools, property…
  • · Adds the right disclaimers (e.g. "I'm not a doctor / lawyer")
  • · Knows which questions must always go to a human
  • · Each playbook bundles those rules + escalation triggers — versioned, so you can audit what changed
📈

It keeps getting better — and proves it

You see the value, not just take our word for it.

  • · Remembers questions it couldn't answer & suggests new help articles
  • · Asks customers if the answer helped (👍 / 👎)
  • · A live scoreboard: how much it handles, how happy people are
  • · Learns from the edits your team makes
🌐

Always on, every channel & language

Covers your customers whenever and however they reach out.

  • · Works on chat, email, WhatsApp and phone
  • · Replies in the customer's own language, automatically
  • · Covers nights & weekends — answers what it can, flags the rest
  • · Greets returning customers like it remembers them

🧩 Want it tailored to your business?

Yes — we'll customize and set it up on your exact docs, tools and stack: your help center, your order system, your industry's rules. Tested and handed over ready to run. That's the done-for-you build (you can also start with a free pilot).

Get it built for you →

Live analytics · this demo's real decision log

It measures itself — provably

Every run above updates these counters live. Your build shows the same on your real tickets: deflection, CSAT, actions taken, and — crucially — the questions it couldn't answer, so you can close the gaps and the agent keeps getting better.

Customer satisfaction (CSAT)

no ratings yet — rate a resolved answer above

The metric every renewal hinges on. Research: AI on routine tickets scores within ~0.2 of human, and the gap closes to near-zero with the human-in-the-loop escalation you saw.

📊 Knowledge gaps it surfaced (questions it couldn't answer → suggested articles)

None yet — try an out-of-scope question above to see the agent flag a gap.

Your numbers

What it's worth to your support team

Every routine ticket the agent grounds and answers is time your team gets back. Set your real volume — the numbers update live.

Median tier-1 deflection is ~41% (Zendesk CX Trends 2026); we model a conservative 30%. Fully-loaded tier-1 tickets cost ~$8; AI resolutions ~$1–2.

$23,400

saved per year

3,600

tickets auto-resolved / year

6 wks

until it has paid for itself

Real-world shape: a team taking 1,000 tickets/month, deflecting ~30%, gets back about 300 tickets a month of human time — and the build pays for itself in roughly 6 weeks. (Illustrative; your free pilot measures yours.)

$0.62 vs $7.40

AI vs human cost per ticket — a ~12× gap (McKinsey, 2026).

$3.50 per $1

Average return on AI customer service; top-quartile 8× (Fin.ai / Gartner).

30% → 67%

Deflection ramps as it learns — ~30% month 1 toward 55–67% by month 6 (Fin.ai benchmark).

Why this one is safe to put in front of customers

A governed agent — not a chatbot that bluffs

01 · GROUNDED

Answers only from your docs

Real RAG over your help center, policies, and past tickets. Every answer cites the source it grounded on. No source, no answer.

02 · GUARDRAILED

Refuses to invent

Below a confidence threshold, or when the docs don't cover it, the agent escalates instead of guessing. The gate is enforced in the workflow, not just asked of the model.

03 · TONE-AWARE

Sensitive tickets go to a human

Anger, threats, refunds disputes, security or data-loss — detected and routed straight to a person, untouched by the bot. No auto-reply on anything high-stakes.

04 · ROUTED WITH CONTEXT

Escalations arrive half-solved

When it hands off, the human gets an AI summary, the customer's intent, the relevant docs, and 2–3 draft replies to pick from — so they finish in seconds, not from scratch. Nothing is auto-sent.

05 · MEASURED

Every decision is logged

Deflection rate, confidence, and outcomes go to a decision log you can audit — so you can prove the ROI and tune what gets auto-answered.

06 · WIRED IN

Lives in your helpdesk

Replies post back to Zendesk / Intercom / Freshdesk / email; escalations land in the right queue or Slack. Built on n8n, on your stack, idempotent and versioned.

That production discipline — idempotency, audit logs, a real escalation gate, evals before launch — is what separates an agent you can trust with customers from a demo that embarrasses you in week two. It's the same checklist we run on every build.

Own it vs rent it

The same governed design — without the forever-bill

The leading AI support platforms are genuinely good. They're also rented: per-resolution fees or enterprise contracts, every year, forever. This agent you own.

What you're comparing This agent Intercom Fin Decagon / Sierra
What it costs $147 once (Core) or $2,497–$4,497 once (built for you) $0.99 per resolution, forever — 3,600 resolutions/yr ≈ $3,564 every year ~$95K+/year enterprise contracts
Who owns it You — it runs on your n8n, your infrastructure Rented — lives on their platform Rented — lives on their platform
Grounded answers · refuses to invent Yes — gate enforced in the workflow, source shown Yes Yes
Audit log of every decision Yes — in your own Postgres, yours to keep On their platform On their platform
Where your customer data lives Your servers — PII masked before the AI sees it Their cloud Their cloud
Best fit SMBs & mid-market who want control without the subscription Teams already on Intercom at scale Enterprises with 6-figure CX budgets

Competitor pricing as publicly reported, June 2026: Fin pricing per fin.ai; Decagon/Sierra contract ranges per published 2026 comparisons. Their products are good — the difference is the ownership model.

Questions, answered straight

Frequently asked questions

What do I need to run the $147 Core agent?

An n8n account (the free self-hosted version works) and an AI key (Claude or GPT). The included setup guide takes about 10 minutes — no coding. You plug in your help docs and it starts answering from them.

What's the difference between the $147 Core agent and the $2,497+ build?

Core is the same governed workflow, set up by you on your own n8n: grounded answers, safe escalation with draft replies, PII redaction. The done-for-you build is us setting it up on your real stack — your help center, your order system, your helpdesk write-back, plus testing on your real tickets before handover.

What happens when it doesn't know the answer?

It says so and hands the ticket to a human — with a summary, the customer's intent, the relevant docs, and 2–3 draft replies to pick from. It never guesses. That gate is enforced in the workflow itself, not just requested of the AI.

Will it make things up, like ChatGPT sometimes does?

No. Every answer must be grounded in a line from your docs, and it shows the line it used. If your docs don't cover the question, it escalates instead of inventing. That refusal-to-invent is the core design, and you watched it in the demo above.

How is this different from Intercom Fin, Decagon or Sierra?

Those are excellent — and you rent them forever. Fin charges $0.99 per resolution with no cap; Decagon and Sierra start around $95K+/year on enterprise contracts. This agent you own outright: it runs on your own n8n, your data stays on your infrastructure, and there are no per-ticket fees, ever.

Is my customer data safe?

Card numbers and personal details are masked before the AI ever sees them. The agent runs on your own infrastructure, not ours, and every decision is written to an audit log you control — which also keeps you ready for compliance reviews.

How long until it pays for itself?

At 1,000 tickets a month with a conservative 30% deflection, roughly 6 weeks — the calculator above uses your real volume. Deflection typically climbs from ~30% in month one toward 55–67% by month six as your knowledge base fills the gaps the agent flags.

What if it doesn't work for my business?

The $147 Core agent has a 30-day money-back guarantee, no questions asked. For the done-for-you build, you can start with a free pilot on one ticket type first — you see real deflection on your real tickets before paying anything.

Get in touch