Guides
A guide is not a chatbot waiting to be asked. It lives on one account with its own memory and a single goal, get this account live, and pursues it for weeks across text, email, and voice. Every answer is grounded in your knowledge base: your docs, your policies, your words. It hands the moment to a human whenever a human converts better.
Portesi’s Pizza, signed to first sale in 13 days. Left: what made Sunny send the message. Right: what the reply changed in the POS, the portal or the plan.
Deal signed in Salesforce. Josh's Airtable tool adds Portesi's to the onboarding queue.
Call held. Specialist sets the disposition in Salesforce, which opens the menu stage.
The guide knows what the menu is missing and asks only for that.
UPS reports the hardware shipped.
Completeness check finds three open questions. Sunny asks one at a time.
Merchant asks something only a specialist can answer. Sunny never promises a callback it can't guarantee.
Specialist approves in the inbox. The guide builds the menu, and a read-back through Tillwell's menu API matches.
UPS: out for delivery, then delivered. Site photos from the install checklist flagged a missing outlet.
Day-before install reminder, by phone.
First card transaction shows up in Tillwell's data, not a status field.
Where it all lands
Every text above is one line in Silvio’s portal.
The plan the guide writes to is the plan the merchant reads: a portal under your brand, magic link, no passwords. Texts, the email and the call all move the same list.
Wed, Oct 7
The three menu questions, answered by text, landed here.
Sunny asked one at a time. The answers went into the draft, Silvio approved the preview, and the specialist got it in their inbox.
Tue, Oct 13
Install day was confirmed the moment all four were green.
Hardware, power, photos and the approved menu. The call the night before wrote to this same page.
An illustrative run. Photos in the thread by brianc and goodrob13 on Flickr, CC BY 2.0.
A chatbot answers whoever shows up. A guide owns an account, remembers everything, and comes back tomorrow.
Anatomy
A goal, not a script. Each guide holds one outcome, module live, record complete, unit open, and decides the next best touch itself. You delegate the goal; you don't choreograph the journey.
Memory that persists. Every touch, reply, and read is remembered for the life of the account. Week three picks up where week one left off, and nothing is re-asked.
Grounded in your knowledge base. Answers come from your docs, your policies, your words, not the open internet. What the guide can't ground, it escalates instead of guessing.
Every channel, one thread. Text, email, and voice share one memory. The customer replies wherever they actually answer, and the guide follows. Voice includes instructions, guardrails, and a number that answers as your brand.
The floor, always on. Consent before the first text, STOP honored instantly, quiet hours in the recipient's timezone, and every dial approved by a human. The full floor →
It can't call its own job done. A guide never marks an outcome complete. The close is a fresh read of your warehouse or CRM, or it doesn't count.
Grounded answers
Guides are assembled from your knowledge base: product docs, pricing policy, setup runbooks, tone. When a customer asks something outside it, the guide doesn't improvise, it checks, or it hands off. That's what makes it safe to put on 5,000 accounts.
Sidekick
And when the next step hides inside the customer's own product — a setting to flip, a snippet to place, a screen they've never opened — the guide sends Sidekick, a white-labeled agent that does the clicking for them: it drives the browser while the customer watches, hands control back wherever a login or an approval needs a human, and the step still closes on your system's read, not on the agent's word.
Start here
The 30-day Verified Activation Pilot: 50–100 customers plus a matched holdout, your brand, your numbers, your consent posture, priced per verified outcome. Build your agent right now and hear it on the phone first.
