Gnews Gnugget Team Responder
Email in, personas out. A listener sends a question to the show's mailbox; minutes later they get a reply written in the voices of Gnorm and Seamus — the same personas from Gnorm's Gnews Gnuggets, arguing it out between themselves. No one is sitting at an inbox making it happen.
It's a working example of what Looptah Connect does in production: take a real inbound channel, put a team of personas behind it, and let it run. Want one voice instead of a panel? That's the Benjamin Franklin Single Responder — same stack, one speaker.
Confidence is provisional. Verification is mandatory.
What it does
One channel in, one channel out — email. Everything in between is handled for you.
Reads the mailbox
Every message that lands in the show's inbox is picked up automatically the moment it arrives — parsed, decoded, and cleaned into plain text ready to work with.
Decides what to do
It checks who sent it, how long it is, and whether it's safe to act on — then routes accordingly. Valid questions go to the personas; everything else gets a helpful, on-brand response or is quietly ignored.
Replies in character
The personas debate the question amongst themselves and the responder emails the result back to the sender — in voice, with the original message quoted below, exactly like a real reply.
Gnorm
Blunt, profane, allergic to hedging. He reacts first with fury, then makes the point land.
Seamus
The humorist. Leads with the joke, lands the argument on the way out — sharper than he lets on.
This example runs a two-persona team, but the same responder can host a whole roster — each with its own point of view and its own job in the reply. Cut it down to one and you have the Single Responder.
A room full of experts, or just one.
Looptah Connect responders come in two shapes, and the difference is not technical — it's editorial. A team responder gives you the argument. A single responder gives you the answer. Pick by what the reader needs.
| Team Responder | Single Responder | |
|---|---|---|
| Example | Gnorm & Seamus (this page) | Benjamin Franklin |
| The reply | Labelled turns — each persona reacts and answers in sequence. | One unlabelled answer, signed by the persona. |
| Best for | Second opinions, red-teaming, editorial and entertainment, anywhere disagreement is the value. | Expert desks, FAQ and support lines, intake, anywhere a clear answer beats a discussion. |
| Cost & latency | One model call per persona per round; richer output, proportionally more of both. | One model call per message — the cheaper, faster of the two. |
| Under the hood | Identical. Same parsing, same allowlist, same anti-spoof and loop guards, same rate limits, same packaging — a single deployment variable decides which one you get. | |
They run side by side quite happily. Each responder gets its own mailbox, its own rate allowance, and its own failure log, so a busy support line never eats into the expert desk's budget — or its reputation.
How it works
The whole thing is event-driven and serverless. Nothing runs until an email arrives. This example is built on AWS — the pattern maps cleanly onto any major cloud.
Mail arrives at SES
Amazon SES receives email for the show mailbox and writes the raw message straight to an S3 bucket. No inbox to poll, no mail server to run.
S3 triggers the function
The new object fires an S3 event that invokes a Lambda function. Compute spins up on demand, only for this message.
The function reads & screens
Lambda parses the MIME email, extracts clean text, and runs the guardrails — sender allowlist, length limits, attachment and anti-spoof checks, and per-sender throttling backed by DynamoDB.
The personas respond
Cleared messages become a task the persona engine runs — Gnorm and Seamus react in turn using the model of your choice, with the API key held in Secrets Manager, never in code.
The reply goes out
The responder composes the reply in persona voice, quotes the original message below it, and sends it back through SES to the original sender.
It goes back to sleep
When the message is handled, the function exits. Nothing is left running and nothing is billed until the next email lands.
No servers to run. No idle bill.
Because every step is triggered by an event, there's no always-on machine waiting for mail. That has real consequences for cost, scale, and operations.
Scales to zero
Between emails, it costs effectively nothing — there's no instance sitting idle. You pay per message handled, not per hour online.
Scales up on its own
A hundred emails at once? The platform runs a hundred copies of the function in parallel and winds back down. No capacity planning, no autoscaling to configure.
Almost nothing to operate
No OS to patch, no server to monitor, no queue to babysit. Managed cloud primitives do the heavy lifting; you own the logic, not the infrastructure.
Hosted by Looptah, or built on your infrastructure.
The responder is a small, portable stack of managed services. That means you can hand it to us entirely — or run the exact same design inside your own cloud account, under your own controls.
Hosted by us
We stand it up in Looptah's cloud and operate it. You point a mailbox at it and start getting replies. The fastest way to a working responder, with nothing for you to maintain.
Built on yours
Deployed into your own AWS account from infrastructure-as-code — SES, S3, Lambda, DynamoDB, and Secrets Manager — inside your VPC, under your IAM, against your data-residency and governance policies.
Like the rest of Looptah Connect, the persona engine runs on the AI platform you prefer — Gemini, Claude, or OpenAI — or a local model when the work can't leave the building. Swap the model without touching the plumbing.
It knows what not to answer.
An open mailbox is an open door. The responder ships with the checks that keep an automated replier from becoming a liability.
Sender rules
A defined list decides who gets a persona reply. Everyone else gets a friendly pointer to subscribe — at most once an hour.
Length & format
Messages that are too long, too short, or carry attachments get a clear note asking the sender to adjust and resend.
Anti-spoof
Spam, malware, and failed-authentication mail is dropped before the personas ever see it — no backscatter, no reputation hit.
Throttling
Per-sender rate limits and loop detection keep cost predictable and stop reply storms cold.
Secrets kept secret
Model API keys live in a managed secrets store with least-privilege access — never in the code or the config.
Failure handling
Anything that errors out lands in a dead-letter queue for review instead of vanishing — so nothing is silently lost.
Want one of these on your own mailbox?
The Gnews Gnugget Team Responder is one example of Looptah Connect. Tell us what channel you'd want a persona team behind — support, sales, intake, or something stranger — and we'll show you what it would take. A whole panel or a single voice. Hosted by us, or built on your cloud.