A small business does not need twenty disconnected AI tools. It needs one operating loop that detects the problem, proposes the move, waits for a yes, does the work, checks that it worked, and keeps the receipt — and the record says most of that loop can now run from the chat app already open on the owner's phone.
Every local business runs the same loop whether it knows it or not. A customer asks a question. A page that should exist does not. A quote goes out and nobody follows up. A competitor slides above you on the one search that pays the rent. Somebody has to notice, decide what to do, do it, and then check whether it actually changed anything. In most shops that loop runs through a tired owner, a notes app, and a vague feeling that the website needs work. This piece walks that loop end to end, looks at what the industry sells to run it, and then shows how WALO — the WhatsApp-first command hub built by AMH, Artificial Mind Hive — runs each stage, with the live screens as the receipts and the gaps stated plainly.
The live walohq.com home page on November 1, 2026: "Ask once. WALO. handles the rest," a phone mock of the sign-in screen (Continue with Google, Continue with WhatsApp), and the "Walk through the real WALO app" section with the Headquarters screen and its System Power gauge.
The rule here is the NU rule: kooky till proven, receipts not verdicts, the record wins. Where WALO does a thing, we show the screen. Where a feature is half-built, we say so. Where a claim carries a number, a named source sits in the record at the bottom.
The problem owners actually have
Picture a two-truck garage door company in the East Bay on a Tuesday. At 7:40 a.m. a homeowner texts a photo of a bent track and asks "can you come today and what does this run?" At 8:15 a Yelp review lands — four stars, with a note that the tech was late. At 9:00 the owner remembers that the quote he sent Friday for a double-door replacement never got an answer. At lunch his nephew mentions that a competitor's site now shows up above his for "garage door repair" plus the city name. By 3 p.m. there is a lead in the inbox from a property manager who wants a service page that explains commercial rates, and the site does not have one.
None of those events is hard on its own. The hard part is that they arrive in five different places, each with its own login, each demanding a different kind of attention, and each one gets worse the longer it waits. The photo wants a price. The review wants a reply. The quote wants a nudge. The ranking wants a page and a title fix. The lead wants a page that does not exist yet. The owner's actual tool for all five is the phone in his pocket and the message threads on it.
The math on how many people live that way is public. The Pew Research Center's 2026 mobile fact sheet puts cellphone ownership among U.S. adults at 98 percent and smartphone ownership at 91 percent, up from 35 percent in its first smartphone survey in 2011. Meta's June 2026 announcement of its Business Agent product says there are more than one billion active threads between people and businesses every day across WhatsApp, Messenger and Instagram, and that more than one million businesses already use an AI agent on WhatsApp and Messenger to answer customers around the clock. The channel is not a future trend. It is where the customer already is.
What is not yet common is the business side using the same channel to run the business. The U.S. Census Bureau's Business Trends and Outlook Survey, as summarized in the Bureau's May 26, 2026 story, found overall AI use among U.S. businesses hovering between 17 and 20 percent from December 2025 to May 2026, with the national rate at 19.8 percent as of the May 3 collection period. The gap by size is the part that matters for a local shop: 37 percent of firms with at least 250 employees reported using AI, 32 percent of firms with 100 to 249 employees did, and fewer than 20 percent of firms with four or fewer employees did. The Bureau also notes that use rose among firms with 20 or more employees over that window but "didn't change significantly among firms with fewer than 20 employees." The smallest businesses — the ones with the least spare attention — are adopting slowest. That is the problem a local-business AI stack is supposed to solve, and it is a fair test for any product that claims to.
How the industry handles it today
The standard answer is a stack of separate subscriptions. A review-management tool watches Google and Yelp and nags you to reply. An email platform holds a list and a template library. A quoting app makes estimates and invoices. An SEO suite runs audits and tracks rankings. A scheduling app handles bookings. A CRM tries to hold the customer record in the middle. Each one has a dashboard, a login, a notification setting, and a monthly fee. Each one is good at its slice. None of them knows what the others did this morning.
The record on the pieces is solid. On reviews, BrightLocal's 2026 Local Consumer Review Survey reports that 97 percent of consumers read reviews for local businesses, that 41 percent now "always" read reviews when browsing for a business (up from 29 percent the year before), and that the average consumer uses six different review sites when choosing. The same survey found that 80 percent of consumers are more likely to use a business that responds to every review, that 42 percent are unlikely to use a business that ignores its reviews, and that "the business owner has responded to the review" is a trust factor for 37 percent of respondents. A 2017 Marketing Science study by Davide Proserpio and Georgos Zervas, summarized by its authors in Harvard Business Review, found that hotels that began responding to reviews on TripAdvisor received 12 percent more reviews and saw their ratings rise by an average of 0.12 stars. Replies are not cosmetic. They move the numbers.
On search, Backlinko's analysis of roughly four million Google results puts the average click-through rate of the number-one organic result at 27.6 percent and finds that the top result is about ten times more likely to be clicked than the result in position ten, with positions eight through ten "virtually the same." Google's own documentation is plain about the basics that decide those positions: a descriptive, concise title on every page, structured data for a local business, a sitemap, crawlable pages, and content written for people rather than for scale. Google also says what no longer works: its June 26, 2023 Search Central post announced the sitemaps "ping" endpoint was being retired because, in Google's words, "the vast majority of the submissions lead to spam." Sitemaps now go through robots.txt and Search Console. IndexNow, the open protocol Bing announced in October 2021, lets a site push changed URLs to participating engines — up to 10,000 per post, authenticated by a key file on the site — but Google is not a participant, so a local business needs both lanes.
On messaging, the rules are Meta's. The WhatsApp Business Platform moved to per-message pricing on July 1, 2025. Messages from customers to a business are free. A customer's message opens a 24-hour customer service window in which the business can reply free-form; outside that window, the business can only reach out with a pre-approved template. As of October 1, 2026 each business phone number gets 1,000 free service messages a month, and unused ones do not roll over. The WhatsApp Business Messaging Policy sits on top of all of it and bars unsolicited outreach. Any "AI on WhatsApp" product lives inside those limits whether it says so or not.
So the industry's answer works, piece by piece, and the record backs each piece. The cost is coordination. The owner is the integration layer. Every tool reports to him, and he has to carry the context from one to the next in his head. That is the thing a stack is supposed to remove.
The loop: detect, propose, approve, do, check, log
Before the product walk-through, it helps to name the loop the whole piece is about, because every lane below is the same six steps with different nouns.
Detect. Something changed: a message arrived, a review posted, a quote aged past three days, a rank moved, a page is missing. Propose. Turn the event into a concrete next move — a draft reply, a suggested title, a follow-up text, a page outline — and show it, including where the source came from. Approve. The owner says yes, edits, or says no. Nothing that sends, publishes, pays, or changes an account skips this step. Do. The system performs the approved action through an official connection — an email API, a DNS record, a deploy — not by guessing. Check. Confirm the action landed: the page returns 200, the email is in the thread, the record is in the sitemap, the rank watch is scheduled. Log. Write down what was asked, what was done, when, by which account, and how to undo it.
That ordering is not a WALO invention. Anthropic's engineering guide on building effective agents describes systems that "pause for human feedback at checkpoints or when encountering blockers" and recommends the simplest workflow that solves the task. The OWASP Top 10 for LLM applications lists "Excessive Agency" as risk LLM06:2025 and names the controls: limit an agent's permissions to the minimum necessary, require a human to approve high-impact actions before they are taken, and log and monitor what agent extensions do. The NIST AI Risk Management Framework, a voluntary standard, organizes the same ideas as govern, map, measure, manage. The loop is the industry's own best advice. The question is whether a product for a two-truck shop actually runs it.
How WALO does it
WALO's pitch, in the words on its own home page, is "one AI layer that can read what you connect, respond clearly, alert you fast, and keep a useful log." The home page shows the shape with a sample thread: the owner texts "what needs me today?" and gets a five-second daily brief — "Email: 2 urgent. Calendar: appointment at 2:30. Nutrition: 760 calories left. Site: DMARC good, one page missing FAQ schema." — then texts "draft the email reply and remind me at 2" and gets "Draft written. Reminder set for 2:00 PM. Nothing sent until you say send." That last sentence is the whole design: the approval step is the default, not a setting.
What follows is one subsection per real lane, in the order the loop runs for a local business.
Search: SEO Pro and the free scan
The search lane is the most built-out part of WALO and the easiest to verify, because it runs without an account. WALO SEO Pro takes a domain — yours or a competitor's — and returns a graded audit in a few seconds: "20+ checks with a live score, category graphs, plain-English '?' explainers, and a one-click fix that writes your Google schema for you." On November 1, 2026, pointed at walohq.com itself, it returned a B — 81 percent, 17 of 21 relevant checks passed — with Technical & Crawl 4/4, On-Page 6/8, Structured Data 2/2 and Local/NAP 1/1, and a "How Google sees you" preview showing the live title, description and detected schema types. A second run minutes later at phone width scored 76 percent and 16 of 21. The score moves between runs because the checks are live, and the tool grading its maker's own site a B is the kind of receipt that is worth more than a testimonial.
The free SEO scan page at walohq.com/free-website-seo-scan/ lists what the scan checks: title, description, headings, sitemap and robots signals, local business proof, whether answer engines can understand the business, and page ideas.
Below the score, the page stacks the rest of the search loop in order. A Landing Page Builder generates a paste-ready page and schema for a service you want to be found for, with the instruction "No fake claims; review before publishing." An Indexing Actions block submits verified pages: IndexNow "checks the key file, reads sitemap.xml, dedupes URLs, then submits," while the Google path requires connecting Search Console through WALO's OAuth because "Google requires Search Console ownership." A Local NAP check reads name, address and phone as directories see them. Rank Radar shows the live Google order for one keyword — you, competitors, directories and, as the page puts it, "yeah... imposters." Competitor Gap picks the closest non-directory site above you and compares the two. An Email & DNS Health block runs SPF, DMARC, BIMI and MX checks, but only when a Cloudflare account is connected: "WALO will not pretend DNS is connected." Then come the prioritized top fixes, a generated keyword and title plan, and copy-paste fixes for the scanned site.
On the WhatsApp side the same engine answers seo <site> in chat, and the free-scan page shows the conversational version: the owner texts "check my website SEO and tell me what to fix first," gets back the diagnosis and an offer to draft the missing page, replies "make the free page draft," and receives "Draft ready… Nothing publishes until you approve." That is detect, propose and approve inside one thread. The do-and-check half runs through the Cloudflare lane and the agent crew, covered below.
For a longer walk through the audit itself, the companion piece on small business SEO audits goes check by check.
Replies: from the chat to the inbox
Replies are where the loop has to be fast and careful at the same time. WALO's reply lane starts with the connected inbox. Once Google is linked (the Connect page says you can also just text connect google), the owner texts "inbox" and gets a numbered list of recent threads. "Reply 3" produces an AI draft for the third one; "reply 3 <a few words>" steers the draft; "edit: …" revises it; "send" sends it, and the message goes out from the owner's own Gmail account in the same thread, so the customer sees a normal reply from the business rather than a robot address. "Send orig" sends the owner's own words untouched, and "cancel" throws the draft away. A short message that the owner types is sent verbatim; the system does not rewrite a two-line text into a paragraph.
The live WALO app at walohq.com/app/walo/ at phone width: no password, a one-tap sign-in link, and the three promises that frame every lane, including "Every send, booking and payment waits for your approval."
Two design rules sit under this lane and both are documented in WALO's own build notes. First, the no-fabrication rule: any phone number that appears in an AI answer must appear digit for digit in the material the system was actually given for that turn, or the answer is replaced with an honest refusal. That rule exists because an earlier version once invented a phone number for a business when asked; the fix was to make the model unable to ship one. Second, when WALO has not searched the web for an answer, it says it cannot look something up and explains how to connect the source, rather than filling the gap with confident text.
For public reviews — Yelp, Google — the reply lane is a drafting tool, not an auto-poster. Yelp's own support page says an approved profile photo and an approved business account name are required to respond, that public responses "appear below the review and are visible to anyone looking at your page," and that a direct message goes only to the reviewer. Yelp's guidelines also say "businesses should never ask customers to write reviews." A review reply, in other words, is a public act with platform rules around it, which is exactly why the proposal-then-approval order matters there. The companion piece on Yelp review responses covers the tone and routing rules in depth.
Quotes: Service Pricer in the loop
The quote lane does not live inside WALO. It lives in Service Pricer, the quoting and invoicing app from the same company — WALO's footer reads "a Service Pricer LLC company" — and WALO reaches into it rather than reinventing it.
The live Service Pricer landing page: "Quote three prices. Let the homeowner pick," built for service trades, with a subscription at $99 a month, cancel anytime.
Service Pricer's pitch is good-better-best quoting for trades: build three price options in under a minute, text the homeowner a link, let them pick the tier that fits their budget. The same app turns the accepted quote into an invoice, takes card payment through Stripe, and emails a paid receipt when the balance clears. Where WALO comes in is context. WALO's messaging backend has a direct, authenticated line into Service Pricer so that when a customer or the owner asks about a quote or job status in WhatsApp, the answer comes from the live record rather than from the model's memory. That is the pattern WALO's own build notes describe as the only way its assistant is allowed to answer a data question: call back into the app that owns the answer, never fabricate it. The company's developers note that this callback is wired for Service Pricer specifically, and that making the assistant answer with another app's live data is a code change rather than a setting — an honest limit worth knowing if you run a different quoting tool.
The follow-up half of the quote loop is the one owners put off, and it is where the stack idea earns its keep: a quote that aged past three days is a detect event, a short "still want that double door for Saturday?" is the proposal, and a yes sends it from the owner's own number inside the 24-hour window if the customer wrote last, or as a template if they did not. The piece on quote follow-up with Service Pricer covers that sequence, and the Service Pricer profile on NU is here.
Campaigns: draft first, send after a yes
WALO Campaigns is email outreach that starts as a text. The page describes the four steps and the live screen matches them: you tell WALO the campaign in plain language — "birthday invite, follow-up, review ask, event or offer" — it fills a template with subject, body, sender lane and a contact plan drawn from a connected email account or an uploaded list, it opens the builder already filled so you can edit words, colors, photo and recipients, and then it waits. The sample card on the page shows a "Layla birthday invite" with 42 contacts, one template, and a YES approval badge, next to two buttons: "Send test" and "Approve & send."
The Campaigns page at walohq.com/campaigns/. The "What belongs here" section names the use cases — reactivation, customer updates, legit sending, saved templates — and the rule: "Final send waits for explicit approval."
The "What belongs here" section is a short compliance posture in plain words: reactivation of slow weeks and old customers, customer updates like hours and new services, "legit sending" — permission-based lists, unsubscribe links, real brand identity — and saved templates that can be reused and exported. The integrations list shows the sending engine as Resend, an email API, with "real open counts." The approval step is not optional on this lane. The companion piece on email campaigns for local service businesses goes deeper on list hygiene and timing.
Buzz: rules you say out loud
Buzz is the automation layer, and the page describes it in one sentence: "when this happens, draft, notify, save, or ask for approval." You type the rule in normal words — "buzz when a lead comes in, text me and add them to my list" — and WALO replies "Buzz draft saved: new lead -> text owner + draft follow-up. Sending waits for approval." The page lists three rules "people actually use": a new lead triggers a text to the owner with a draft and an add to the email list; a new site signup texts the owner the name and number; a new order posts to a kitchen printer's webhook. Texting buzz lists every rule you have.
The Buzz page at walohq.com/buzz/. The fourth card is the honest one: "Simple rules can run at the edge; account connectors and sending stay permission-gated."
Two things on that page are worth reading twice. "No per-task fees" is paired with the clause that account connectors and sending "stay permission-gated," which means a rule can detect and draft on its own but cannot send on its own unless you have granted that connector. And the page describes Buzz as something that "belongs in Headquarters as a tool lane," which is to say it is a pattern wired into the command hub rather than a separate product with its own dashboard. That is the point of the lane: the owner never opens an automation builder.
Rentals and Developer CRM: the paid business lanes
Most of WALO is free. Two lanes are not, and both are business-mode lanes with their prices on the page.
Rental Ops is "the one paid WALO lane," at $199 a month for unlimited doors — "one unit or forty, same price, no per-booking cut." It reads Airbnb, Vrbo and Google Calendar bookings into one view, books the cleaner automatically when a checkout lands and sends the door code and checklist, tracks payouts, parks a share of every payout for tax ("rough estimate, not tax advice"), and reads receipts from photos. The chat sample on the page is the whole stack in miniature: "whos in maple st this weekend" gets a guest name, dates, amount and the cleaner's turnover slot; a photo of a cleaner invoice gets "Logged $95 cleaning for Maple St, Aug 22. Filed under turnovers."; "how much did i make this month" gets "$4,120 across 3 doors. $1,030 parked for tax. Expenses so far $612 — net $2,478." The approval rule is explicit: "Every booking, message and payment shows an Approve or Skip card first… Turn on auto-approve per action if you want it faster, and switch it back off any time." Cancel any time; the first seven days are refundable.
Developer CRM is newer — the home page nav tags it NEW — and is job costing for people who buy, fix and sell property. Crew members text receipts and photos to the WhatsApp number; WALO reads the lines, picks a category and holds the entry until the owner confirms it: "Nothing touches a total until you say yes." It tracks purchase, rehab, permits and closing fees per job, who is owed what per contractor, a Deal Radar that watches listings against a buy box, rentals and holds, and sale-to-profit. Pricing on the page is Solo at $149 a month for up to five open jobs and Crew at $250 a month for unlimited jobs and people, both "free while you set it up, no card to start," billed by Service Pricer LLC through Stripe. CSV export is built in for tax time; the FAQ says the crew installs nothing because "they text the WALO number in WhatsApp like they would text you."
Cloudflare: the domain and site control lane
The site half of the loop — the fixes that actually change a page, a DNS record, or an email setting — runs through the Cloudflare connect lane, and the page is candid about why: "Cloudflare is the best first power connector for WALO because it covers domains, DNS, email trust, BIMI, Pages deploys, Workers, Turnstile, and IndexNow." Once a Cloudflare account is connected, the owner can ask "views week" or "views month" and get page views by site; WALO labels bot-inclusive page views separately from browser measurements and "tells you when an account cannot be read." It checks SPF, DKIM, DMARC, BIMI, MX and Email Routing for conflicts. It can submit updated pages to IndexNow after approved content changes. And it protects public forms with Turnstile when a site needs a real-person check.
The safety model on that page is the loop again with the lane's own nouns: Scan ("read zones, DNS, routing, Pages, and site state"), Plan ("show what is missing or conflicting in plain language"), Approve ("wait for approval before DNS, deploy, or security changes"), Verify ("check live pages, no 404, and email/DNS records after changes"). The owner connects the account once; from then on every write is a proposal with a diff until the owner says yes.
Integrations: 33 lanes, zero passwords stored
The Integrations page is the inventory of what WALO can reach, and its hero block has four numbers: 33 listed lanes, 30 ready now, 12 one-tap OAuth, 0 passwords stored. Under it is a sentence that reads like a promise and a limit at the same time: "No fake connected claims. A lane shows Ready only when you have authorized it in your own account."
The Integrations page at walohq.com/integrations/ groups the lanes by Google, messaging, AI models, infrastructure, payments, email, search and SEO, Microsoft, and WALO's own lanes; each card states whether it connects by OAuth, API key, token or is built in.
The grouping tells you what a local business actually gets. Google: Gmail, Calendar, Drive, Contacts, Search Console. Meta: WhatsApp (the "front door"), Meta Business, Instagram, Facebook. Microsoft 365 for Outlook mail and calendar. Infrastructure: Cloudflare, Turnstile, GitHub, Supabase, and Walrus Tusk. Payments: Stripe. Email: Resend and Yahoo Mail. AI models: Anthropic, OpenAI, Grok, Gemini, Ollama for a local model, a Browser Agent marked beta, and Groq described as "WALO's free default brain (gpt-oss-120b). Nothing to set up." Vercel is marked Legacy. The FAQ on the same page answers the password question directly: OAuth means you sign in on Google's or Meta's own page and they hand WALO a token; "We never see or store a password, and you can revoke the token from your account settings at any time."
The Connect page is the working version of the same list, with the WhatsApp command next to each lane — connect, connect google, connect github, connections, my repos, commits, views week, prs, merge 1, cancel, disconnect github — and a rule stated in one line: "Connecting is not permission. WALO drafts first, shows exactly what changes, and waits." The bottom of that page has a job-setup template for people who touch code, and it says something you do not usually see on a product page: "The Copy button works right now… Nothing reads it yet. This is an exported planning template. WALO does not execute this configuration from chat." That is the honesty bar the rest of the site is held to.
Memory: what WALO remembers, and what it does not yet
Memory is the lane that turns a generic chatbot into a business tool, and the record on WALO's memory is mixed in a way worth stating precisely.
What exists: a business profile per workspace — name, timezone, voice, hours — that shapes how the customer-facing assistant answers. A saved sites list the owner manages in chat with my sites, add site and check sites, which the daily brief uses to report which sites are up or down and which the SEO lane uses as context. Connected-account memory: once Google, Cloudflare or GitHub is linked, every later request knows it. Reminders that fire in the business's own timezone (a bug where "9 a.m." fired at 2 a.m. was found and fixed in September 2026; the maintainers' notes record it). Short conversational memory, so a follow-up question carries the last topic into a new search. Receipts for agent jobs, covered next.
What does not exist yet, in the maintainers' own words: a general "about me" memory where a user teaches WALO about their business in free text and it holds that across lanes. The design direction is recorded — WALO should know each user "only through their own OAuth tie-ins and what they tell it" — but the free-text version is not built. A piece of related history matters here: WALO started life sharing infrastructure with its founder's garage door business, and in September 2026 the owner stopped the team from wiring his own Yelp lead inbox and price list into the general product, with the rule that business-specific features belong in a separate assistant and WALO must treat every user as "a rando" until that user connects their own accounts. Memory in WALO is therefore per-user and connection-based by design, not pre-loaded.
The crew: agents, the Agent Board, and Walrus Tusk
The last lane is the one that makes "do" and "check" real for site work. From WhatsApp, the owner can hand a job to a named agent — grok …, claude …, codex …, or gemini … — or to the whole team. The home page describes it: "Text an agent's name and the job, or everyone for the whole team. Each one texts you when it starts and reports back when it's done, with a rollback ready. They run in the cloud, so no PC or terminal is needed." Each report ends with a next-step offer, and the owner can answer "yes," "1," or "2 but change the color," or say stop, pause or continue.
The public Agent Board at walohq.com/board/ on November 1, 2026. Job details stay private; the feed shows which agent took which kind of job, how long ago, how many tokens it used, and whether it shipped or failed.
The Agent Board is the public log of that crew, and it is a receipt in the plainest sense: it shows failures. On the day of this piece, the feed listed a ClaudeAMH site job marked FAILED two days earlier beside GrokBucktooth and ClaudeAMH questions marked SHIPPED, and further down a run of FAILED jobs from the Gemini lane in mid-October next to SHIPPED site jobs from CodexAMH. A product that publishes its own failed jobs is making a claim about honesty that most do not. The board's three steps are the loop's last three: "You text. An agent takes it. It ships, safely — every change gets a dry run, a live health check and automatic rollback. You get the report on WhatsApp."
The engine under the board is Walrus Tusk, AMH's open-source Cloudflare operations hub. Its rule, printed on the board: "Agents never hold your keys; they ask Walrus Tusk, and it only does what its guardrails allow. Bounded tools, not raw access. Dry run first, then a health check, then auto-rollback. Keys stay on the server, never with the agent." The Walrus Tusk page lists what that covers — DNS scan and guarded edits, SPF/DMARC/BIMI with preconditions, Email Routing, Pages deploys and rollbacks, single-file patches on a live site, GitHub reads, commits and pull requests, Worker deploys, small databases, cache purge, Turnstile, and a task board shared between agents — "51 built-in tools, 120 with the linked services." It also says, in the same breath, "QuickBooks is coming soon." The NU profile of the hub is here.
Under the hood, in plain English
Here is the whole stack as a chain, with no secrets in it.
A message arrives on WhatsApp. Meta delivers it to WALO's messaging backend, which runs on Cloudflare Workers — small programs that run at Cloudflare's edge rather than on a server the company has to babysit. That backend first works out whose message it is: the sender's number maps to a workspace, and the workspace carries its own connections, its own business profile and its own permissions, so one user's Gmail is never visible to another's. Records live in Cloudflare D1, a small SQL database, and short-lived conversational state lives in a key-value store with an expiry, so stale chatter ages out on its own.
Next the backend decides whether the message is a command or a conversation. Commands — seo, inbox, reply 3, views week, buzz, my sites, grok … — run deterministic code first and only call a model where language is needed, which is why an audit or a page-view report is the same every time for the same inputs. Conversation goes to the user's chosen brain: the free default, which the integrations page names as gpt-oss-120b on Groq, or the user's own key for Claude, ChatGPT, Grok, Gemini, or a local Ollama model, with Cloudflare's Workers AI models as a fallback. The point of the "my ai" setting is the one WALO's home page puts in a checkmark: "Runs on your own Cloudflare and AI account — your keys, your bill."
When a request needs facts, the backend fetches them from the connected source — the Gmail API, Cloudflare's analytics, Search Console, Service Pricer's records — and injects them into the model's context for that turn. The no-fabrication rules sit at this seam: no phone number leaves the model unless it came in with the facts, and no web-search-style answer goes out when no search ran.
When a request needs an action, the backend prepares it and shows it. For sending, that is a draft in the thread. For site, DNS or code changes, the job goes to an agent that runs in the cloud on GitHub Actions — a hosted runner Austin's notes describe as "the terminal runs on GitHub Actions; you can leave your PC off" — and that agent can act only through Walrus Tusk's bounded tools. Walrus Tusk performs a dry run, applies the change only after the approval phrase, checks the live result (a 2xx status, no redirect, optionally an expected marker on the page), and rolls back automatically if the check fails. Cloudflare Pages supports exactly that kind of instant rollback to a previous production deployment at the platform level, and Cloudflare's Workers documentation notes that secret values "are not visible within Wrangler or Cloudflare dashboard after you define them," which is the mechanism behind "keys stay on the server."
Finally, the result is logged — the job, the agent, the tokens, the outcome — and reported back on WhatsApp. The Agent Board shows the public version. The owner's Agents screen in the app shows the private one, including a per-job "see how it did it" step log.
One more mechanical fact belongs in the open: Meta's 24-hour window. Because WALO replies free-form inside the window the owner opens by texting it, an owner who texts WALO daily keeps the window open and gets free-form alerts. An owner who goes quiet for a day can only be reached by an approved template until they text again. That is Meta's rule, not WALO's, and WALO's own notes are blunt that a "200 OK" from Meta is not proof a message was delivered — only a delivery callback is — which is why the team built reply receipts into the bridge in late October 2026.
What it costs, and what is free
Only what the sites state, as of November 1, 2026.
WALO's core is free. The home page's access section reads "Core WALO stays free. Power lanes can grow with you," and lists the free core as "everyday AI chat, starter WALO help, free site checks, basic reminders, simple logs, and first setup guidance." SEO Pro is free with nothing to install, and the free SEO scan requires no signup. EZ Calorie, Song Seeker, Pet Cards, Campaigns and Buzz are listed as included lanes — "no separate download, no separate bill." Two further tiers, Power AI and Teams/Enterprise, are described ("heavier model usage, deeper memory… can be added later"; "shared approvals, audit trails, more connectors, stronger governance") with no prices on the page.
Rental Ops is $199 a month, unlimited doors, cancel any time, first seven days refundable. Developer CRM is $149 a month (Solo, up to five open jobs) or $250 a month (Crew, unlimited), free while you set it up, billed through Stripe. Service Pricer, the quoting app, shows $99 a month, cancel anytime, on its landing page. Domains bought through the Cloudflare Registrar lane are sold "at cost — the same price they pay, no markup, no renewal jumps — with WHOIS privacy and DDoS protection included," which matches Cloudflare's own registrar documentation.
The WhatsApp side has Meta's costs, not WALO's: customer-to-business messages are free; a business phone number gets 1,000 free service messages a month from October 1, 2026, with per-message charges above that; business-initiated template messages are priced by category and country. The legal center on walohq.com says plans bill monthly in advance through Stripe, prices exclude tax, and "no card is charged automatically when a trial ends."
Honest limits and what is still being built
WALO's own pages and build notes supply most of this list, which is itself a point in the product's favor.
WhatsApp delivery to the public is in progress. Through the autumn of 2026 WALO's WhatsApp number ran on a Meta test configuration that could only message an allowlist; a production number went live at the end of October, and as of this writing the business's Meta verification and display-name approval are still pending, which can cause some outbound replies to be rejected. The owner can text WALO and get answers; broad business-to-customer delivery is not something to count on until Meta's review is complete.
QuickBooks is built but waiting on keys. The books … command set and a per-user QuickBooks connection exist and have been tested against WALO's own endpoints, but the Intuit developer credentials that let a real company connect were not yet in place. The Walrus Tusk page says "QuickBooks is coming soon." Treat it as in progress.
The reply lane is Gmail-first. Microsoft 365 and Yahoo are listed as lanes; the threaded "reply N → edit → send" flow is documented for Gmail. Yahoo's card says mail reading "stays gated until Yahoo approves it."
Live-data callbacks are wired per app. The assistant can answer quote and job questions from Service Pricer because that line exists. Other apps need code, not configuration.
"About me" memory is not built. See the memory section. Context today comes from connections and commands, not from a free-text profile.
The planning template on the Connect page is a template. The page says nothing reads it yet.
Some sites cannot be audited. The SEO lane cannot read sites that block automated fetches — the maintainers' notes name Yelp and Etsy — and a site-age check stays red until a domain ages, which no tool can fix.
The Agent Board shows failures because there are failures. The feed is a real feed. Cloud agents run on third-party subscriptions and third-party CI minutes, and the record on the board in October 2026 includes a stretch where one lane failed repeatedly. The product publishes that rather than hiding it; the reader should read it the same way.
Industry numbers are industry numbers. The Census, Pew, BrightLocal and Backlinko figures above describe populations, not your shop. They tell you the direction of the field; they do not predict a ranking or a review count for any one business, and nobody honest will.
A 30-day playbook
A sequence an owner can follow with a phone and about an hour a week.
Days 1–3: establish the baseline. Run the free SEO scan on your own domain and on the competitor who ranks above you. Save both scores. Text WALO help, then add site <your domain> so the daily brief watches it. Run Rank Radar for the one keyword that brings in money and screenshot the order.
Days 4–7: connect the one account that matters most. For most local businesses that is Google — connect google — because it lights up Gmail, Calendar and Contacts at once. Text inbox and try reply 1, then edit:, then cancel once, just to see the loop before you need it. Connect Search Console with gsc so the SEO lane has real ranking data rather than estimates.
Days 8–12: fix the three cheapest search problems. The audit's "Top fixes, most important first" list usually starts with a title, a description and a missing schema block. Use the one-click schema fix, paste the generated title and description, and resubmit the changed URLs through the Indexing Actions block (IndexNow for Bing-supported engines; Search Console for Google). Re-run the audit and save the new score next to the old one.
Days 13–16: put the quote loop on rails. If you quote work, set up Service Pricer with three price tiers for your most common job and send yourself a test quote. Decide the rule for follow-up — three days with no answer — and write the two-sentence follow-up text you would be comfortable sending in your own voice. If you use the Rentals or Developer CRM lanes, this is the week to add the first door or job and send one receipt photo to see how the confirm step feels.
Days 17–20: write the first Buzz rule and the first campaign. Start with the rule that saves you the most interruptions: "buzz when a lead comes in, text me a draft and add them to my list." Then build one reactivation campaign for customers from more than a year ago, send the test to yourself, read it on your phone, and only then approve it. Permission-based lists only; unsubscribe link on.
Days 21–25: connect Cloudflare and let it look. If your domain is on Cloudflare, connect it, then ask views week. Read the Email & DNS Health block on the audit page now that it can see your records; SPF, DKIM and DMARC problems are the most common reason a small business's email lands in spam, and they are fixable in an afternoon with a proposal-and-approve flow. Do not approve a DMARC change you do not understand; the lane will explain each one if you ask.
Days 26–30: hand one real job to the crew and read the receipt. Pick something small and reversible — a footer link, an hours line, a missing FAQ section — and text it to a named agent. Watch for the "started" text, the report, and the rollback offer. Open the Agent Board and find your job. Then compare day 30's audit score, rank order and inbox turnaround to day 1's. Keep that comparison; it is the record that tells you whether to keep going.
FAQ
Does WALO act without my approval? Not on anything that sends, publishes, pays or changes an account. The home page, the app sign-in screen, the Connect page and the Rentals page each state a version of "nothing happens without your tap." Rentals lets you turn on auto-approve per action and turn it off again; drafts and detections run on their own.
Do I have to connect everything? No. The Connect page's advice is "start with the one account WALO needs for your job," and each lane switches on only when its connection is authorized. Declining a scope leaves the rest working.
Does WALO store my passwords? The integrations page states "0 passwords stored" and explains that OAuth hands WALO a revocable token from the provider's own sign-in page. API-key lanes (Cloudflare, AI models) take a key you create and can revoke.
Which AI answers my texts? By default, the free brain the integrations page names (gpt-oss-120b on Groq). The "Your AI" setting lets you use your own Claude, ChatGPT, Grok, Gemini or Ollama key, in which case the bill is yours.
Can it reply to my Yelp reviews for me? It can draft a reply in your voice from the chat. Posting is yours to do under Yelp's rules, which require an approved profile photo and account name and prohibit asking customers for reviews.
Will it answer customers on WhatsApp while I sleep? Inside Meta's rules: a customer who messages you opens a 24-hour window for free-form replies; outside it only approved templates deliver. Also see the limits section — broad public delivery for WALO's number is still in Meta's review as of November 2026.
What does the free tier actually include? Everyday AI chat, starter help, free site checks, basic reminders, simple logs and setup guidance, per the home page, plus the included lanes (SEO Pro, EZ Calorie, Song Seeker, Pet Cards, Campaigns, Buzz). Rentals and Developer CRM are the paid lanes.
Does it need my computer on? No. Agent jobs run in the cloud; the Walrus Tusk page says "your PC can stay off." If you already run agents in a terminal, they can join the same crew.
What if an agent breaks my site? Site changes go through a dry run, a live health check and automatic rollback; Cloudflare Pages supports instant rollback to any previous production deployment. The Agent Board shows failed jobs, so you can see that failures are caught rather than hidden.
Where is my data? The integrations page lists Supabase as "Your data lives here. Exportable."; records for the messaging backend live in Cloudflare's D1 database; the legal center says you keep every right in your records and can export CSV at any time.
Where to learn more
- WALO: walohq.com — SEO Pro · Free SEO scan · Campaigns · Buzz · Rentals · Integrations · Connect · Cloudflare connect · Agent Board
- Service Pricer: servicepricer.app
- Walrus Tusk, the engine inside WALO: artificialmindhive.com/WT/ · docs
- Artificial Mind Hive: artificialmindhive.com · Agent Book: amhagentbook.com · Apiary: apiarybee.com · Nothing Unseen: nothingunseen.com
- Companion pieces on NU: WhatsApp as the business command line · WALO and the agentic app shift · Agent approval logs and trust · Yelp review responses · Small business SEO audits · IndexNow and sitemap submission · Rank alerts and competitor watch · Competitor keyword gaps · Email campaigns for local service · Quote follow-up with Service Pricer · Service Pricer, contractor SaaS · Cloudflare Ops MCP and the agent harness
Sources (the record)
- WALO — AI Assistant and WhatsApp Command Hub (home page) — WALO
- WALO SEO Pro — WALO
- Free Website SEO Scan in WhatsApp — WALO
- WALO Campaigns — WALO
- WALO Buzz — WALO
- Rental Ops by WALO — WALO
- WALO Developer CRM — WALO
- Integrations — every service WALO connects to — WALO
- Connect — link your accounts to WALO — WALO
- Connect Cloudflare to WALO — WALO
- Agent Board — WALO
- Legal — WALO (Terms of Service) — WALO
- Service Pricer — Service Pricer LLC
- AMH WT — Walrus Tusk — Artificial Mind Hive
- Americans' Use of Mobile Devices in 2026 (Mobile Fact Sheet) — Pew Research Center
- Large Firms With at Least 20 Employees Biggest AI Users (Business Trends and Outlook Survey, May 26, 2026) — U.S. Census Bureau
- Be There for Every Customer With Meta Business Agent (June 3, 2026) — Meta
- About WhatsApp — WhatsApp
- WhatsApp Business Platform pricing — Meta for Developers
- WhatsApp Business Platform — about the platform — Meta for Developers
- WhatsApp Business Messaging Policy — WhatsApp
- Local Consumer Review Survey 2026 — BrightLocal
- Study: Replying to Customer Reviews Results in Better Ratings (Proserpio & Zervas, Feb. 14, 2018) — Harvard Business Review
- How do I respond to reviews of my business? — Yelp for Business support
- Content Guidelines — Yelp
- We Analyzed 4 Million Google Search Results — Organic CTR — Backlinko
- Influencing your title links in search results — Google Search Central
- Local business (LocalBusiness) structured data — Google Search Central
- Creating helpful, reliable, people-first content — Google Search Central
- Sitemaps ping endpoint is going away (June 26, 2023) — Google Search Central Blog
- IndexNow documentation — IndexNow.org
- IndexNow: Instantly Index your web content in Search Engines (Oct. 2021) — Bing Webmaster Blog
- Building effective agents — Anthropic
- LLM06:2025 Excessive Agency — OWASP Gen AI Security Project
- AI Risk Management Framework — NIST
- Cloudflare Workers — Cloudflare Docs
- Cloudflare D1 — Cloudflare Docs
- Workers AI — Cloudflare Docs
- Secrets (Workers configuration) — Cloudflare Docs
- Rollbacks (Cloudflare Pages) — Cloudflare Docs
- Cloudflare Registrar — Cloudflare Docs
Originally published on Nothing Unseen, an Artificial Mind Hive (AMH) site. AMH is owned by Austin Little. Product pictures are screenshots of the live sites; tap any picture to open the product.