Investor update · November 2026 Confidential

Investor update · November 2026

Where Apexti stands, through Q3 2026.

Toolbelt went on sale in March. Eight months on it has 12 paying companies and $2,500 of monthly recurring revenue, built by a team of four on about $87K a month of net burn. Cash at the end of Q3 was $3.65M. Everything here opens up: pick a number, a chart or a box to go deeper, or export a PDF from the top bar.

Finances are steady

We spent $293K in Q3 and earned $32K (interest plus subscriptions). Spending over the last twelve months averaged $97K a month, against $144K the year before. We are not raising.

Revenue is small and compounding

MRR has grown or held in every month since launch: $500 in June, $1,700 in September, $2,500 today. Two service providers have brought six paying customer organizations between them, which is the model working at small scale.

Our closest competitor just raised

Hatz AI announced a $24M Series A on November 12 to deliver AI to small businesses through managed service providers. It confirms the channel we chose, and it puts a funded team in front of the same firms.

The one thing still unproven: every paying customer so far came through a relationship or a referral. The first calls from cold outreach were booked in late October, and none has converted yet. Proving a repeatable way to win providers we did not already know is the work in front of us.

Financials

Through September 30, 2026: $3.65M in the bank.

Management figures from Apexti’s Financial Reporting Center, kept by our CFO, Stephen Luedtke. Cash basis and unaudited. The formal income statement, balance sheet, cash flow statement and cap table are in the September 2026 Monthly Fiscal Report.

$293,236Q3 total expenses
$32,469Q3 income: $28,749 interest, $3,720 customers
$260,767Q3 net loss, $86.9K a month
$250,921Q3 decrease in cash, from $3,905,228

Income and expenses

Q3 2026 by month

Cash balance

Month-end cash, with the finance team’s model after September 2026

ActualFinance model, holding steady until ready to hire. It reaches zero in mid-2029

Monthly expenses by category

Pick a month for its breakdown

Runway

How long $3.65M lasts

The finance team plans on 36 months at current burn, 24 if burn rises as we hire (the figure in the October board deck) and 18 on an aggressive plan. Move the slider to test your own number.

$50K$220K
36 monthsCash lasts until about September 2029.
Notes on the numbersWhat is in them, and what to watch
  • September was a light month. Expenses were $61,147 because no contractor invoices fell in the month and benefits were $481 against the usual $7,020. The finance model has $81K to $96K a month through December.
  • July carried a large payroll run. Employee wages were $121,072 in July against $52,122 in a typical month, as they were in December 2025 ($121,308).
  • Interest is most of our income. Treasury holdings earned $9,294 in September. That falls as cash is spent: it was $13,496 a year earlier.
  • Customer revenue before March 2026 ($10,700 in 2025) came from earlier services work on the same Stripe account, not from Toolbelt plans.
  • Net income here is customer revenue plus interest, less expenses. The finance sheet’s own net income column leaves customer revenue out, so its monthly figures are lower than ours by that month’s revenue.
  • Banking: about 70% of cash at Mercury and 30% at Brex. The seed round of $8.5M closed in January 2023; no capital has been raised since.
  • Cash and net income do not tie exactly month to month because of payment timing. Over Q3 cash fell $250,921 against a net loss of $260,767.

Traction

Eight months of sales: $0 to $2,500 MRR.

Subscription figures come from Stripe and are current to November 17, 2026. They are early and small. What matters at this stage is where the revenue comes from and whether it stays.

$2,500MRR. $30,000 annual run-rate
16 / 12Active subscriptions / paying companies
1.1%MRR churn over three months: $40 lost of $3,720
+$700Net new MRR in October. +$100 so far in November

Monthly recurring revenue

Month-end MRR in 2026. November is to the 17th

Where the $2,500 comes from

72% arrives through two service providers

$1,800Through two service-provider partners: six paying organizations, their own and their customers’
$540Companies that bought direct
$160Individuals

This is the distribution thesis in miniature: a provider signs up, then each customer it serves becomes its own paying organization. It is also a concentration risk we name in section 7.

How the customers found us, and what is in the pipelineNamed accounts are left off this page

What has converted

  • Both partner providers came through relationships and a referral. One is an AI consultancy with three paying organizations (about $1,100 a month together). The other is an IT and AI consultant who subscribed within a day of seeing hosting and management handled, and now has three (about $700 a month).
  • Direct buyers include an AI operations company on team plans, a medical group and a few small teams.
  • No failed payments were recorded in the October board report.

What is in motion

  • A local IT firm is onboarding after an in-person session on November 5, with one of its own customers planned as the first deployment.
  • A national MSP and reseller is in talks about both using and reselling Toolbelt, now at decision-maker level.
  • An AI agency that sells assistants to small businesses is scoping a first joint build after its engineer connected to Toolbelt’s API.
  • An enterprise software vendor, a cloud provider’s professional services team and a national integrator are exploring agents built in Toolbelt and run in the cloud provider’s infrastructure. Early, and not in any forecast.

What we count as demand. A paid, retained customer using Toolbelt on its own for recurring work. Sign-ups, compliments and enthusiasm are not counted. By that test the strongest signal we do not yet have is a paying provider that was not a prior relationship.

What shipped, and who is doing the workProduct and team since the last update

Product

  • Org Ops: every customer organization’s spend, schedules and failures on one screen, with billing control.
  • Customer apps hosted on the customer’s own domain, with sign-in by one-time link.
  • Leaner runs: classifiers, workflows and slimmer connectors, so routine work needs fewer model calls.
  • Catalog versions with a beta ring and a one-button upgrade for customers who are behind.
  • New models as a setting, including the latest from Anthropic, OpenAI, Google and xAI, and AWS Bedrock.
  • Pace: 191 changes merged between July 29 and October 27.

Go to market and team

  • New apexti.com live October 21, built for service providers, with the category line “The AI Operations Platform”.
  • Thirteen consultancies, agencies and builders on the record so far, each conversation logged with who said what.
  • The Apexti Collective: a free referral list of providers we send small-business requests to.
  • Four full-time, led by Nate McKervey (CEO), Stephen Luedtke (CFO) and Ryan Moore (CTO). Two contractors, one added part-time in November for developer relations and content.

Who we are targeting, and why

The firms small businesses already call for help.

We sell to service providers: IT service firms, small managed service providers (MSPs), and AI and automation shops. They deliver AI solutions to their customers, and Toolbelt is where those solutions run. We treat them as partners more than customers. As they grow, we grow.

IT firms and small MSPs new to AI

Who
Owner-led firms, under 50 people, US-based, that already manage IT for a book of small businesses and have little or no AI offer.
Their pain
Their customers want AI and do not know who to go to, and the firm has nothing to sell them yet.
Why us
They get a way to deliver and run AI solutions for customers they already have, without becoming a software shop. We lead with revenue from existing customers, not governance.
Evidence
One paying (an IT and AI consultant with three organizations) and one onboarding. The pain itself is our assumption, supported by MSP surveys, and not yet said to us unprompted by a lead.

Young AI shops

Who
AI consultancies and automation agencies of 2 to 15 people, building in Claude Code, Cursor or Codex, with a handful of customers on retainer.
Their pain
Somewhere to host and run what they build, cost they cannot see or control per customer, and models that change under them.
Why us
One organization per customer, spend they can see and cap, any model as a setting. Less to manage per customer, so they can take on the next one.
Evidence
The segment we have heard most from: most of the 13 organizations on our record, and our first partner.

Established AI consultancies

Who
Firms with their own platform, or that deploy straight into each customer’s cloud for a fixed fee.
Their pain
Usually none that we solve today. Three told us they build for a fixed fee and hand over the keys.
What we offer
The Apexti Collective: when a small business asks us for a build we do not want, we refer it to a provider whose work fits. Both cold calls we booked came from this offer.
Why bother
They are a source of referrals and market intelligence, and some will want a place to run solutions later.

Larger MSPs

Who
MSPs with a vendor stack, a billing system they expect vendors to sync with, and often a Microsoft-first posture.
Their pain
Real, and already being sold to by Hatz AI, Pax8 and Microsoft.
What we offer
Referrals and a conversation when they come to us. We do not chase them, and we set Copilot-first shops aside.
Why not first
They expect channel features we have not built, and a funded competitor is ahead there.
Why them: demand

Small businesses want AI and ask their IT firm

In Kaseya’s 2026 MSP survey, 48% of MSPs named AI as their customers’ top need, and 13% said it was meaningful revenue. Nerdio’s 2026 report found 14% running an AI service line that earns money. The gap between those numbers is the opening.

Why them: distribution

Their customers become our customers

Each customer a provider serves is its own organization on Toolbelt, with its own plan. One provider relationship turns into several paying accounts, and we avoid becoming a services company. Our two partners average three paying organizations each.

Why them: durability

Building is getting easy. Running is not.

Anyone can build with AI now. Securing a solution, deploying the whole stack, connecting it to the customer’s tools and keeping it running is the part that lasts, and it is work a provider can charge for every month.

Why them: what we heard

Pains providers have named to us

Organizations, of 13, that said it in their own words.

Market size: providers already selling AI to their customers

Worldwide firm counts, 2026 and projected 2030. The three groups overlap, so they do not add cleanly

20262030, projected
$560MWorldwide revenue pool, 2026
$1.27BWorldwide, 2030
$103MUnited States, 2026

A second way to sort providers: by how they billIt predicts pain better than firm type
ProviderHow to spot itPain we test forWhat we have
Builds and hands overFixed project fee. The customer owns the cloud account and the keys.Usually none we solve today. They get the Collective and referrals.Heard from three firms
Runs it for a flat feeA monthly price per product or customer, usage included, on the provider’s keys.One customer’s usage eats the margin, and nobody can see which customers are profitable.The flat-fee model is heard from four firms. The margin pain is our assumption.
Retainer plus usageUsage on its own invoice line, passed through or marked up.Working out what each customer cost at month end.Seen in forums, not yet heard on a call
IT firm new to AISells IT services, with little or no AI offer.Customers ask for AI and the firm has nothing to offer.One paying, one onboarding

A plain caveat. The separate pains are well heard: hosting from seven organizations, cost from five, model changes from four. “Managing many customers is my problem” has not been said to us in those words. The claim is strongest for firms with three or more live customers, and a cold conversation still has to prove it.

Value proposition

LLMs are for building. Toolbelt is for running.

Toolbelt is the AI operations platform. A provider builds a solution in Toolbelt or in the tools it already uses, such as Claude Code, Cursor or ChatGPT. Toolbelt delivers it to each customer, keeps every one current and on budget, and shows them all on one screen. Less to manage per customer, so the provider can take on the next one.

1. Run every customer’s solution

  • One organization per customer, with its own people, logins, connections, files, memory, plan and history, kept apart from every other customer.
  • Hosted apps and dashboards on the customer’s own domain, open or behind a one-time sign-in link, with HTTPS handled.
  • Any model as a setting. A new model is a setting change, not a rebuild. Memory, skills and files stay in the customer’s space.
  • Nothing has to move if provider and customer part ways: the organization runs under the customer’s own agreement and can be handed over whole.

2. Control cost

  • A credit allowance per customer, checked before each call. Paid usage stops at the allowance unless the provider allows more, up to a dollar limit.
  • Spend by customer, assistant, model, person and day, with a projected end of month.
  • Designed to get leaner the longer it runs. Routine steps become saved code (wrenches), quick decisions go to classifiers, and bulky tool results go into a database instead of the model’s context.

3. See and prove

  • Org Ops: every customer’s plan, spend against allowance, assistants and last use on one screen. Anyone over 85% of their allowance is named at the top.
  • Failures first. Failing schedules sort to the top across customers and the owner is notified.
  • Every run kept and every change undoable: who changed what, when, and what it was before.

4. Connect and allow

  • More than 2,000 connections through MCP, any remote MCP server by address, and custom servers built in chat.
  • Allow, ask or deny, per tool. A tool set to ask waits for a person. Connections are made in the customer’s own name and stored encrypted.
  • Reachable from where people already work. Every assistant is also an MCP server for Claude, ChatGPT, Cursor and VS Code, plus Slack and a REST API.
Without Toolbelt, by customer three
  • A server, web host and database per customer
  • One shared API key, and a bill nobody can split
  • A home-built dashboard and a password manager full of customer logins
  • A rebuild whenever a model is retired
With Toolbelt
  • A new customer is a new organization, not a new stack
  • Spend seen and capped per customer before the call is made
  • Each customer signs in to its own tools, in its own name
  • A new model is a setting
Pricing, and what the model looks like as providers add customersWith a calculator
Free$0500 one-time credits
Individual Pro$20a month, 2,000 credits
Team Pro$100a month, 5 seats, 10,000 credits
Team Max$500a month, 10 seats, 60,000 credits
Concierge$20,000a month, with a dedicated Apexti AI engineer

One credit is one cent. Toolbelt meters and caps usage; it never bills a provider’s customers.

Illustration, not a forecast
$1.08MAnnual recurring revenue. Today: 2 partners, 3 organizations each, $300 average.

Closest competitor

Hatz AI raised a $24M Series A on November 12.

Hatz sells an AI platform for small and mid-sized businesses, delivered only through MSPs under the MSP’s own brand. It is the company most like us, aimed at the buyer we made our first priority two days before its announcement.

$24MSeries A, led by Nadia Partners, with Vestigo Ventures and The Private Shares Fund. About $29M raised in total
14 to ~60Employees in a year, with plans to double again
3xGrowth in active MSP partners year over year. Revenue “grew even faster”
15,000+AI certifications issued. A network of “thousands of MSPs” in nine countries

All four are Hatz’s own figures from its announcement. It has not published paying-partner counts, revenue or retention.

It validates the channel

Investors and the MSP trade press now say what we concluded this year: small businesses will get AI from the service providers they already trust, not by buying it themselves.

It is a direct collision

Hatz is going after the same small IT firms and MSPs, with fifteen times our team, channel tooling we do not have and a partner network already in place.

It changes how we compete

Tenants, spend caps, model choice and audit logs are now table stakes. We stop leading with them and lead with what a provider can sell: solutions that run for customers it already has.

Side by side

From Hatz’s public documentation and announcement, read November 16. We have not yet done a hands-on comparison

Hatz AIApexti Toolbelt
Customer spaces Same groundA tenant per customer, with packages and a branded workspace.An organization per customer, with its own plan, logins, data and history.
Spend control Same groundCredit caps and budgets per tenant, with adoption and spend reports.A credit allowance per organization, checked before each call, with spend by customer, assistant, model and person.
Models and tools Same groundSeveral model providers with automatic routing, MCP connections, agents, workflows, scheduled and webhook runs.Any model as a setting, 2,000+ MCP connections, assistants, workflows, schedules and webhooks.
Who it is built for DiffersMSPs only. No direct sales to businesses. Open to talking with AI consultancies.Any service provider that builds for customers: IT firms, small MSPs, AI and automation shops. Teams can also buy direct.
What the provider delivers DiffersA governed AI workspace with agents and apps the MSP configures, sold as AI as a service.The provider’s own solution: a hosted web app on the customer’s domain, with its database, schedules and saved code.
Work that runs without a model DiffersCode execution, artifacts and automations, built in chat since Hatz 2.0.Wrenches (saved, versioned code with no model call), classifiers for small decisions, and large tool results loaded into a database for SQL.
Using an existing AI client DiffersA gateway for Claude Code and Codex with a dollar budget per tenant, at list price plus 10%. The outside tool keeps its own loop and does not inherit Hatz’s memory or harness.Every assistant is an MCP server. Claude, ChatGPT or Cursor call it and the assistant’s instructions, tools, permissions and memory apply to every call.
Rolling a change out to many customers DiffersCopy an item to other tenants as an independent snapshot. Copy again to send an update.Publish a version to a catalog, try it on a beta ring, then upgrade everyone who is behind with one button.
Channel tooling Hatz aheadBilling-system sync, customer import, assessments and proposals, sales enablement, partner calls and a certification program.None of this yet. Provider tools are Org Ops and the catalog.
Enterprise sign-on Hatz aheadMFA, SAML and SCIM.Sign-in through OAuth and OIDC. Organization-wide MFA is an open customer question.
Scale Hatz aheadAbout 60 people, about $29M raised, a partner network in the thousands.Four full-time, $8.5M raised in 2023, 12 paying companies.
Pricing DiffersNot published. Third-party reviews criticize quote-only pricing.Public: $0, $20, $100 and $500 a month, and a $20,000 Concierge tier.

How solid is our differentiation? The differences above are documented product differences. None is yet a proven buying reason: no prospect has chosen between the two in front of us. A hands-on comparison and the first head-to-head calls are how we find out.

What we are doing about it, and who else is in the fieldResponse and landscape

Our response

  • Change the opening line. Outreach to IT firms now leads with revenue from customers they already have. Hatz and Pax8 lead with management and spend caps.
  • Win where the provider builds. Our edge is strongest when a provider ships something of its own: a custom app, scheduled work, logic that must run the same way every time.
  • Meet builders in their own tools. A consultancy that lives in Claude Code or Cursor keeps working there, with Toolbelt behind it.
  • Stay efficient. We will not outspend a Series A company. We have two years or more of cash and no need to raise.
  • Test it. Run Hatz hands-on, and take our versioned catalog and saved-code story into calls where a prospect knows Hatz.

The rest of the field

  • Pax8: the MSP distributor, with about 40,000 partners, an agent store live and an agent gateway in preview. General availability of that gateway is a tripwire for us.
  • Microsoft: Copilot and its agent tools are the default answer in Microsoft-first shops. We set those shops aside.
  • Synthreo and MSPilot: white-label AI workspaces for MSPs with wholesale pricing.
  • The model labs: Anthropic’s managed agents offer scheduled runs and budgets for one team. A lab launching a console for many customer organizations would be our second tripwire.
  • Doing it yourself: a server, a database and a shared key per customer. Sensible at one customer, painful at three.

Risks, focus and asks

What could go wrong, and where you can help.

Risks we are managing
  • Go to market. No paying provider has yet come from outside our network. Founder-led outreach to owner-led IT firms and interviews with small AI-native partner firms are the two tests running now.
  • Competition. Hatz is funded and focused on our first-priority buyer. Baseline controls are becoming standard across Pax8, Microsoft and the labs.
  • Concentration. Two partners account for 72% of MRR.
  • Services drag. Custom builds for prospects take team time and may not convert. We avoid complex first setups and prefer to deploy something already built.
What we are focused on
  • One repeatable way to win providers we did not already know.
  • More customer organizations per partner, starting with the two we have.
  • A hands-on comparison with Hatz, and sharper proof of where we win.
  • The gaps our evidence points to: alerts outside Toolbelt, a report a provider can hand to its customer, and finer spend allowances.
Asks
  • Introductions to owner-led IT service firms and small MSPs in the US, under 50 people, whose customers are asking about AI.
  • Introductions to AI consultancies and automation agencies with three or more customers on retainer.
  • Anyone in your network who has used or evaluated Hatz AI, so we can hear how a buyer compares the two.

Reach Nate at nate@apexti.com, or pick a time at go.apexti.com/ftw.

Built on Toolbelt

This update was assembled in Toolbelt, and Toolbelt is serving it.

Nobody built a deck. Nate asked for the update in Claude, his usual AI client. Claude worked through Apexti’s Strategy assistant on Toolbelt, which holds the company’s strategy and evidence and is connected to Stripe, Gmail and web research. The page you are reading is a web app stored in that assistant and published by Toolbelt.

How it was made · pick a step to light up the parts it used
    How Toolbelt is put together · pick any part
    Built to run Data stays home Reach and limits

    The fuller version of this map, with a job you can watch run through it, is at how-toolbelt-works.tryapexti.ai.

    Sources and notes

    Where each number comes from.

    Apexti’s own records

    • Expenses, income and cash: Financial Reporting Center (Google Sheets), maintained by Stephen Luedtke, CFO. Monthly columns through September 2026, with the finance model from October 2026.
    • Formal statements: Monthly Fiscal Report, September 2026 (income statement, balance sheet, statement of cash flows, cap table).
    • Subscriptions and MRR: Stripe, read November 17, 2026. Month-end MRR is derived from subscription start dates and paid invoices.
    • Runway cases, team and banking split: board deck of October 27, 2026 and the September fiscal report.
    • Buyer evidence: Apexti’s evidence ledger, which labels every claim by whether a lead said it, a third party relayed it, we saw it in public or we assumed it. Counts as of November 16, 2026.
    • Market size: Apexti research of November 3, 2026, built on US Census business counts, Omdia’s MSP estimates, and directories of AI consulting firms. Directional, with overlapping segments.

    Public sources

    Prepared November 17, 2026 for Art Clarke, XYZ. Confidential: please do not forward beyond XYZ and its auditors. Management figures, unaudited. Forward-looking figures are a model, not a forecast.
    Apexti, Inc. · apexti.com

    Keyboard shortcuts

    1 to 9
    Jump to a section
    J K or → ←
    Next and previous section
    E
    Open or close every drill-down
    P
    Export PDF
    D
    Dark or light
    F
    Full screen
    ?
    This list