iPhone or Android: Set Up Your AI Teammate Where You Chat

Configuring a Hey Button AI teammate needs no installation on your iPhone or Android, because you simply direct it to the business channel your customers already utilize, such as a business phone number that receives texts, a business email inbox, or a messaging integration, and it works there.

Why doesn't an AI teammate need its own app?

It doesn't need one, because an app helps only when someone remembers to open it, and a customer texting your business number has no reason to know that a separate app exists. The most useful place for a teammate is the channel where the request already arrives, which is why Hey Button works through that channel directly.

Under the hood, Hey Button operates on OpenClaw, an open automation framework, with Hermes as part of the underlying runtime, and models such as Claude and GPT answer through it depending on the task. It is not a proprietary model built from scratch. None of that software operates on your phone, which serves only as the place where your team reads the results.

Consider this illustrative example from a lawn care business. A customer texts, "Can you give me a quote for a spring cleanup?" The teammate organizes the inquiry, notes that the address and photos are missing, and drafts a reply for approval, so nothing is sent until someone approves it. The owner sees the request in the place they already verify, and no new icon appears anywhere.

The trade-off is fit. If your staff already work inside a scheduling or job-management tool, the teammate must operate alongside that tool through the channels it can reach, adding a layer on top of your existing work rather than replacing your tools. For a broader view of the role, read what an AI teammate is.

What does setup look like on an iPhone?

On an iPhone, setup is a decision about which business channel the teammate monitors, not a permissions tour of your personal device. You are not granting access to your personal Messages app or your Photos library, because the teammate operates only on the business channel you name during setup, and the rest of the phone remains outside its scope.

Setup proceeds through three stages, each with a clearly defined completion point:

• Stitch is an intake conversation covering how work arrives, which tools you already utilize, and where repetitive tasks accumulate. It is complete when the teammate's workflow matches how requests genuinely arrive at your business.
• Press is go-live, during which you establish explicit approval rules governing what the teammate may read, draft, or perform independently, and which items always return to you first. You observe it handling real examples before anything reaches a customer, and the stage is complete when those examples align with your judgment.
• Mend is ongoing tuning, supported by a client portal that demonstrates what the teammate is doing and why. People on your team direct that tuning, rather than the teammate adjusting itself.

For an iPhone owner, the practical sequence unfolds as follows:

1. Determine the business number or inbox where customer requests land.
2. Complete the Stitch conversation and confirm which channel is in scope.
3. Choose an access level for each kind of task.
4. Review the Press examples and sign off on the approval rules.

For how these pieces fit into the wider setup flow, see how Hey Button works. The trade-off is that the majority of the effort lies in Stitch and Press rather than on the phone, because a rushed approval rule either forces you to review too much or allows through something you wanted to examine first. The iPhone is the straightforward part.

What does setup look like on Android?

On Android, setup follows the same pattern, with the teammate connecting to the business number or inbox rather than to the phone itself, so nothing needs installing on the device you carry. Because Android supports a wider range of messaging apps, confirm during Stitch which specific channel should be monitored, whether that is SMS, a business email account, a form, or a messaging integration.

Consider a cleaning business as an example. The owner answers booking texts from a dedicated business number on an Android phone that also holds personal conversations, and during Stitch the central decision is which number falls within scope. The teammate monitors that number and nothing else on the device. When business and personal conversations share one thread or login, the boundary becomes harder to explain, so resolve the separation before go-live, because separate numbers or accounts make that line easier to draw.

The trade-off is that Android setups frequently involve more app choices, and therefore more decisions up front, so recording those choices during Stitch prevents rework later in Press.

Which channel should the teammate observe first?

Begin with the channel where unanswered customer requests already accumulate. Connecting every channel on day one introduces approval rules you have not tested and makes it harder to determine what the teammate handled. Because each channel needs its own rules, one thoroughly tested channel is worth more than three partially configured ones.

Treat the table as a starting point rather than a verdict. If two channels carry the same kind of request, begin with the one where customers wait longest for a reply.

What should I gather before the Stitch conversation?

Collect the channels, tools, and requests you already handle, because Stitch builds from what happens today. A teammate configured around real workflows needs real examples rather than a description of how you wish work arrived.

Bring the following items:

• The business phone number, inbox addresses, and any form or messaging account customers utilize.
• The tools your team already utilizes for scheduling, quotes, or invoices, if any.
• Three to five recent customer requests, each paired with the reply you genuinely sent.
• A short list of decisions that always need a person, such as pricing exceptions or refunds.

Each business archetype has its own version of this list:

• Lawn care: estimate requests texted during the week, seasonal cleanup inquiries, and schedule changes for mowing.
• Contractor: estimate requests by email, job-site photos sent by customers, and follow-ups on quotes that have not been signed.
• Cleaning business: booking changes, access instructions for homes, and rescheduling texts.
• Small law office: intake inquiries through a form or email, appointment confirmations, and document request follow-ups. Anything that sounds like legal advice is a human-only item you establish in your rules.

The trade-off is scope creep at the start, so do not attempt to gather everything, because a lengthy list slows Stitch down. Begin with the requests that recur most frequently, and add the remainder when they appear in Press.

Can the AI teammate see my personal texts and photos?

No. It can see only the business channel you connect during setup, such as a business phone number, a business email address, or a specific messaging integration, and it has no access to your personal device, your contacts beyond that channel, or your photo library. Anything you have not directed it toward remains outside its reach.

Scope and access are separate settings, since scope determines what the teammate can see at all while the access level determines what it may do with what it sees. A small law office, for example, might allow the teammate to draft intake replies while keeping spending and sensitive changes behind approval. Granting access for one task does not widen the channel scope for every other task.

The trade-off is that scope is only as clean as the channel you hand over, because a shared phone that mixes business and family messages in one thread makes the boundary difficult to explain to your team. Settle that question during Stitch, before anything goes live.

What access level fits each kind of task?

Select the lowest access level that eliminates the busywork for each task, and raise it only after Press examples demonstrate that the teammate performs reliably. Levels are assigned per task rather than through one global switch, so a single teammate can operate at different levels for different work.

Draft-and-approve is a sensible starting point for most small teams, because the review step reveals what the teammate gets right before you extend it more latitude. The trade-off with full access is that each step up grants the teammate more room to act, which means your rules must be more precise to match.

How do I test the teammate before customers see it?

Test the teammate during Press by observing it handle real examples you select, before any customer message is sent. The test is complete when its drafts and decisions match what you would have done yourself.

Conduct the test in four passes:

1. Select five to ten past requests across every channel in scope, including at least one difficult case.
2. Operate each request through the teammate and compare its draft to the reply you genuinely sent.
3. Classify each result as ready, needing a rule change, or must always return to you.
4. Update the approval rules wherever the drafts missed, then repeat the pass until the labels stabilize.

The trade-off is time, since Press needs patience. Skipping it shifts the first real mistakes onto customers, which is precisely what approval rules exist to prevent. AI can make mistakes, so this review forms part of the setup rather than an optional extra.

What if my team utilizes a mix of iPhones and Android phones?

That arrangement works, because the teammate lives on the channel rather than on any single device. An office manager on an Android tablet and an owner on an iPhone observe the same teammate working the same business channel, and neither needs to install anything to see what it did.

Consider a contractor whose owner verifies the business inbox from an iPhone while on job sites, and whose office manager reviews drafts from an Android tablet at the desk. Both see the same drafts awaiting approval, and neither device needs any added software.

The trade-off is that notification settings live on each person's device, so each person configures how they want to be alerted. Agree on who owns approvals before go-live, because if two people can approve the same draft, it may eventually be approved twice or not at all.

Will my phone get flooded with notifications?

Notification volume follows the access level and approval rules you select during setup, so you determine how much reaches you. Under draft-and-approve, you are notified only about items requiring a decision, such as a drafted reply, a quote to review, or a spending item held back, and you receive no running commentary on every message the teammate reads.

Customer messages, spending, and sensitive changes remain behind the approval rules you establish, which makes them the items most likely to reach you, and that is the intended purpose of those rules. Routine items your rules allow the teammate to handle independently need no attention and should generate no alerts.

The trade-off operates in both directions: loose rules mean fewer alerts and greater risk, while strict rules mean more reviews and slower replies to customers. Press is where you find that balance by observing real examples, and Mend adjusts it later, with the rules still set by people.

How do I keep it working after go-live?

Review what the teammate did against your rules on a regular schedule, and modify a rule only when the review reveals a gap. Mend is where that work takes place, through ongoing tuning supported by a client portal that demonstrates what the teammate is doing and why.

A simple weekly verify serves most small teams well:

1. Open the client portal and scan the requests handled since the last review.
2. Mark anything that went against your intent, whether a draft missed a detail or an item should have come back to you.
3. Adjust one rule at a time, and record the reason beside it.
4. Repeat the process the following week and observe whether the same gap recurs.

The trade-off is that changing many rules at once makes it difficult to determine which change helped, whereas small, dated adjustments remain easier to trust later. The gains come from your review rather than from the teammate tuning itself.

When does a phone-based setup not fit?

It doesn't fit when the work lives somewhere the teammate cannot reach. Hey Button teammates operate through text, email, and forms, and they do not handle phone calls. A law office that takes most intake by phone keeps those calls with people, then utilizes the teammate for written follow-up, such as the confirmation email or the intake form, once that channel is connected.

It also doesn't fit when no one owns approvals, because drafts left unreviewed slow customers down more than handling the work by hand would. Before go-live, designate the person who reviews drafts and the backup who covers when that person is unavailable.

The trade-off is straightforward: a phone-based setup fits when requests arrive by text, email, or forms and someone is ready to review them. If those pieces are missing, fix the process first, since no channel setting can repair it.

Frequently Asked Questions

Do I need to install an app on my iPhone or Android to utilize an AI teammate? No — nothing has to be installed on the phone you carry. The teammate connects to the business phone number, business email inbox, or messaging channel you choose during setup, and works there. Your team can verify results from whatever device they already utilize, including a mix of iPhones and Android phones.

Can an AI teammate read my personal texts or photos? No — it only sees the business channel you connect during setup, such as a business phone number, a business email address, or a specific messaging integration. It has no access to your personal device, your contacts beyond that channel, or your photo library. Anything beyond that needs your separate approval.

How is pricing handled for an AI teammate? Typically, setup and ongoing support are quoted around your workflows and integrations. Any software subscription or usage charge is identified before you commit, with no surprise line items. Ask for that breakdown during setup, before you agree to anything.

What happens when a request is missing details? Typically it lists what's missing, such as the address or photos, and drafts a reply asking for those details. Under draft-and-approve, nothing is sent until someone approves it. A quote request with missing details is the kind of gap Press examples are meant to catch early.

Does the teammate need software installed on my computers? No — the software does not operate on your phone or on your team's computers. You connect the business channel, set the approval rules, and review results where you already work. Setup begins with the Stitch conversation, not a software install.

Ready to set this up? Start the setup questionnaire at heybutton.ai/start and tell us which channel your customers already utilize.