On October 7, 2026, TechCrunch reported that Tab emerged from stealth at a $300 million valuation, calling it "another personal AI assistant" in a crowded field that includes Meta's Muse and others. The pitch is simple: you text it on iMessage or WhatsApp, and it does things for you, from grocery orders to birthday gift ideas, using what its co-founder describes as its own phone number, computer and wallet.
This post covers what has been reported, what it implies about how Tab works, how its trust claims compare with similar assistants, and what to ask before you let any assistant handle your errands and money. A sourcing note: we saw excerpts of the TechCrunch article and secondary coverage that repeats them, not the full article or Tab's own materials, so details are marked where we could not confirm them.
TL;DR: the questions people are asking
| Question | Short answer |
|---|---|
| What is Tab? | A personal AI assistant you text on iMessage or WhatsApp |
| Who is behind it? | Co-founders Brennan Erbz, Stafford Schlitt and Ammar Amdani (spelling of the first name unconfirmed) |
| Valuation and funding? | $300M valuation; amount raised not disclosed |
| Investors reported? | SV Angel, Valar Ventures, American Spirit |
| What can it do? | Grocery orders, gift ideas and other errands, per the founders |
| How is it different? | Has its own phone number, computer and wallet, so it acts for you |
| Privacy promises? | No training on user data; payment vault Tab cannot see; approval before irreversible actions |
| Is it available? | Mostly word of mouth; a waitlist or private beta is reported but unverified |
| Is it the Tab wearable? | Not as far as we can tell; see the name-collision note below |
What was announced
TechCrunch's report places Tab in the same category as other consumer assistants that handle reservations, shopping and travel. Co-founder Ammar Amdani says the product is meant to act, not advise: it "has its own phone number, computer and wallet," so it can carry out a request on your behalf instead of telling you how to do it.
Reported details:
- Interface. You interact by messaging, on iMessage or WhatsApp, rather than through a new app. That lowers friction, because the assistant lives where you already talk to people.
- Team. Three co-founders who began the work earlier this year.
- Backers. SV Angel, Valar Ventures and American Spirit are reported as investors.
- Valuation. $300 million, with the amount raised undisclosed.
- Growth. Mostly word of mouth so far.
- Roadmap. The company says it will launch a new product soon and did not say what.
A name-collision note. There is an older, unrelated product also called Tab, a wearable AI pendant created by Avi Schiffmann and covered in 2024. Third-party databases list it as a small, earlier-stage company. We found no evidence the two are connected, so do not conflate them when you search.
How it probably works, and what we are inferring
Amdani's "own phone number, computer and wallet" line is the most interesting technical claim, and we want to separate what he said from what we infer.
Own phone number. Messaging through iMessage and WhatsApp implies Tab is reachable at its own contact, and that it can also place calls or receive verification texts as itself. That is a different model from an assistant that reads your inbox as you.
Own computer. Most agent products run a browser or virtual machine on the user's behalf. An "own computer" suggests Tab operates in its own environment, which might be a cloud machine, to click through websites. This resembles the architecture of cloud-VM agents we have covered, such as Meta Muse.
Own wallet. An assistant with a funded wallet or a payment instrument it controls can buy things without you typing card details into each site. Amdani says payment details sit in a vault Tab itself cannot see, which would mean the agent triggers payments without reading the card number. How that is implemented, for example through tokenized cards or a payments provider, was not described in the coverage we reviewed.
Our inference, flagged as such. Acting from its own accounts rather than borrowing your logins could sidestep one concern retailers raised against other agents, credential capture. Amazon's complaint against Meta's Muse included that it appeared to store customer credentials, as we covered in Amazon blocks Meta's Muse. But Amazon also objected that Muse did not identify itself as an agent. Whether Tab identifies itself to the stores it uses, and whether retailers accept that, has not been reported. It is the same question Paul Graham raised about agents and Amazon.
The trust pitch: three promises, three ways to check them
Amdani frames trust as the main differentiator. The three promises are specific enough to test.
| Promise | What would verify it |
|---|---|
| User data is never used to train AI | A written policy, contract terms and, ideally, a third-party attestation |
| Payment details sit in a vault Tab cannot see | A description of the payments architecture, a security assessment and a named payments partner |
| Nothing irreversible without explicit approval | A list of what counts as irreversible, how approval works in a text thread, and logs of every action |
None of these has been independently confirmed in the coverage we saw. They are the right promises, because the most-reported failures in this category involve approval. Our write-up of ChatGPT Dots sending an unapproved email shows why "explicit approval" has to be precise: an assistant that interprets a casual message as consent can still act wrongly.
How Tab compares with other personal assistants
We have covered several assistants in this category. The comparison below uses only facts from our earlier posts and the Tab report; blanks mean we do not know.
| Tab | Meta Muse | Hark Pro | Underdog | OpenClaw | |
|---|---|---|---|---|---|
| Where you use it | iMessage, WhatsApp | Meta app and devices | Web, iOS, Android (US) | Mac app | Self-hosted, chat apps |
| Where it runs | Not stated; implies its own computer | Cloud VM | Cloud, with a computer-use model | Local on Apple silicon | Your machine or server |
| Money handling | Own wallet, vault Tab cannot see (claimed) | Not covered here | Encrypted credential vault (reported) | Browser actions gated by human approval | You configure it |
| Pricing | Not published | Not covered here | Reported Free, $20 and $100 tiers | Not covered here | Open source |
| Weights or code | Closed | Closed | Closed | Local models | Open source |
For fuller context, see Hark Pro versus Muse and Dots, the Underdog local assistant, the OpenClaw guide and OpenAI's Dots.
Why a text interface matters
Putting an agent in iMessage and WhatsApp is a distribution bet. People already have the apps open, group chats make it easy to share an assistant, and there is nothing to install. It also has costs.
- Weaker confirmation controls. A chat thread is a poor place for structured approvals. "Yes" can mean many things, and a long thread hides what was approved.
- Platform dependence. Messaging platforms set rules for automated accounts, and those rules can change.
- Shared threads. In a group chat, who is allowed to instruct the assistant, and with whose money?
- Auditability. A chat history is a log, but not a clear ledger of what the agent did.
A good design would add structured confirmations, spending limits, per-merchant rules and a readable activity log, even inside a text interface.
What to ask before you use it
If you join a waitlist or get access, these are the questions worth answering first.
- What exactly can it spend, and how do I cap it? Ask for per-purchase and monthly limits.
- What counts as irreversible, and how is approval requested? Look for a distinct confirmation, not a casual "ok."
- Where does my data live, and who can read the thread? Including Tab staff.
- Is anything used for training, including by model vendors it calls? "We do not train" and "our vendors do not train" are different statements.
- How does it identify itself to retailers and services? This affects whether it keeps working.
- What happens if it is wrong? Refunds, liability and dispute handling.
- Can I export or delete everything? Test it early.
What this means for what you build
If you build assistants, Tab's reported design is a pattern: separate identity for the agent, a payment path that hides card data from the model, and messaging as the interface. The unresolved parts, approval UX, merchant acceptance and auditability, are where products in this category will win or lose. See our guide to human-in-the-loop design and the agentic commerce stack from Stripe and others for the building blocks.
If you are a user, treat a $300 million valuation as a statement about investor expectations, not about safety or quality. Start with low-stakes tasks and a small spending cap, and wait for independent reviews.
Related reading on explainx.ai
- Hark Pro: Brett Adcock's proactive assistant vs Muse and Dots
- Underdog: a private local personal AI
- Is Meta Muse safe? The honest verdict
- Amazon blocks Meta's Muse from shopping
- Paul Graham: Amazon banning agents is a startup opportunity
- ChatGPT Dots and the unapproved email
- What is OpenClaw? Personal AI assistant guide
- Agentic commerce: Stripe, Link and Grok shopping
Details are based on TechCrunch's October 7, 2026 report as summarized in excerpts and secondary coverage, and on statements by Tab's co-founders; they may change. We have not used the product or reviewed its policies. Valuation and funding are not the same thing, and this is not investment advice.