1. Investment thesis
EXME turns an AI agent’s capabilities into a service for people who want to delegate work without setting up servers, models or integrations. Users describe a task in a familiar chat and receive research, a document, a calculation or a prepared action. The platform combines models, tools, memory and a workspace, and maintains them as a service.
The investment thesis rests on recurring useful work. If an assistant reduces the user’s effort and delivers a usable result, the product has a basis for a subscription. If delivery costs remain below revenue after support and acquisition, repeat use can support a sustainable business.
The technical foundation is in place: an execution engine, environment management, memory, tool connections, model selection and cost accounting. The next task is to turn these components into a repeatable user experience and validate economics with paying cohorts. Commercial metrics need to be established through pilot results and subsequent sales.
The key investment checkpoints are user acceptance of results, repeat tasks, conversion to payment, retention and contribution after all direct costs. Financial scenarios show the assumptions under which the model works and expose sensitivity to execution quality, price, CAC and churn.
2. The problem and market opportunity
A powerful model may have a simple interface, but completed work usually needs more: obtaining materials, choosing tools, granting access, keeping context and checking the result. Users must coordinate separate applications or assemble their own automation.
This friction is especially visible in recurring document and information work. Each additional service brings setup, limitations and learning. Maintenance continues after the first launch: changing integrations, request costs, lost context and error handling.
EXME’s opportunity is to become a convenient place to delegate that work. Customers pay for access to a managed assistant and less coordination effort. Demand for this service must be established through frequency of use and willingness to renew.
3. Initial segment and market expansion
The initial segment is independent professionals and owners of small projects who regularly prepare proposals, reports, research and client correspondence. They already use digital tools, can judge the result and can decide to buy a subscription themselves. This creates a short path from task to purchase and early feedback.
The first product focus is a limited set of frequent tasks with verifiable outcomes. After repeat use is established, expansion can include small teams, industry packages and additional personal scenarios. Team roles and organizational requirements are part of the development plan.
The reachable market should be estimated from the bottom up: audience in accessible channels × share with suitable tasks × conversion to payment × revenue per customer. Initial channel and cohort results will refine the reachable customer count and annual segment revenue. Scaling follows measured demand rather than general interest in AI.
4. Customer value and a strategy for advantage
EXME’s promise is: “Tell us what you need. EXME gets to work.” The product combines five elements: an accessible entry point, execution through tools, context continuity, control of access and spending, and environment maintenance. Users do not need to become administrators of their assistant.
Alternatives include general-purpose AI chats, specialized applications, automation builders and self-assembled agents. EXME competes on time to a useful result, ease of repeat requests and total cost of use. Advantage must be demonstrated on concrete tasks in the initial segment.
The strategy for defensibility is to accumulate verified workflows, reliable integrations, knowledge of common failures and durable customer relationships. Memory and saved workflows can make continued use more valuable; trust requires understandable data practices and user control.
Access to external models and open protocols does not itself create exclusivity. EXME’s potential advantage is the quality of the full service, from the first connection to an accepted result and predictable cost. This is a business-building strategy, not a claim of an established technology monopoly.
An important direction is to turn recurring requests from nontechnical users into maintained automations without manually building a workflow graph. Combinations of AI and deterministic steps already exist, including in n8n. EXME’s focus is to connect these approaches to a personal assistant, user context and the transition from a successful request to repeat execution. This capability is planned and must demonstrate its usability and economic effect.
5. Work customers may pay for
Research for a decision. Compare options against a budget, timeline and constraints. The result is a table, conclusions and sources the user can check.
A document for real work. Turn customer materials into a proposal, report or presentation. Value is measured by how usable the file is and how much editing it still needs.
A prepared action. Draft an email or a record change in a connected service. Users grant access and approve consequential actions; the assistant reduces preparation work.
Project continuity. Return to materials, agreements and preferences without explaining the full context again. Important conditions are clarified when stored information is out of date.
These are priority product scenarios. Their commercial scope depends on the readiness of specific tools and integrations. Quality means a result the user accepts, rather than answer length or the number of steps executed.
6. Architecture: how a request gets done
EXME has three layers. The interface receives requests and presents results. The management layer handles users, access, environments and spending. The agent engine chooses a sequence of actions, calls models and tools, and preserves context.
The task path is request → skill instructions → model selection → a tool call through a plugin → verification → result. In practice, the loop can repeat: reading a document may reveal the need to refine a conclusion, obtain more data or ask the user to decide.
A skill explains how to do the work. A plugin provides access to a particular tool or API. The router allocates tasks to models with different costs and capabilities. These separate roles allow workflow quality, model choice and integrations to improve independently.
The assistant receives a separate workspace for files and tool execution, with resource limits. Managed updates, state recovery and cost control turn technical components into a maintained service. Isolation reduces risk but does not imply absolute security.
EXME develops the product and operating layer on top of agent technologies and external LLMs. The strategy does not require building a foundation model: resources focus on user outcomes, execution quality and service economics.
Chain crystallization into pipelines
The next architectural stage is “baking validated action chains.” An agent solves a task and checks the result; a successful, repeatable sequence is then captured as an executable pipeline. A normal run follows validated steps without calling an LLM for every step. New conditions, ambiguity and exceptions return to the agent for review.
This can turn everyday requests into complex automations: reading data, transformations, calculations and API calls follow a defined sequence, while a model is used where reasoning is needed. This is a planned mechanism, not a description of an already available feature.
Repeat execution requires input validation, chain versions, limited permissions and result checks. Changes to an external service should stop execution or trigger a review of the affected step. The approach may substantially reduce LLM computation; API calls, infrastructure, support and exception handling still have costs.
7. Ecosystem: memory, hundreds of plugins and thousands of skills
Memory helps retain useful summaries, preferences and project context. It makes repeat work easier and reduces the need to start every task from scratch. Important information needs freshness checks; managing history and access is part of product quality.
Hundreds of plugins and thousands of skills out of the box are EXME’s planned offering. This describes the target scale of a future catalog, not a verified count available today. Priority areas are email, documents, CRM, analytics and development.
A plugin connects the assistant to an external system. A skill provides a ready-made workflow, such as researching a question, assembling a document or checking a result. Together they expand the work users can delegate through one interface. Personal services connect with the user’s permission.
Catalog expansion should follow demand. Priority goes to integrations that increase useful usage and retention. Each needs a clear scenario, access and failure checks, maintenance and acceptable delivery cost. Catalog breadth becomes an advantage only when quality holds.
8. Model strategy
The engine is separate from the LLM provider. This allows economical models to handle routine work and more resource-intensive models to handle complex tasks without changing the user’s interface. Choice also reduces dependence on one provider.
| Family or direction | Role and status |
|---|---|
| GLM | Connection is part of the current technical foundation; a direction for economical execution. |
| Claude | Supported through compatible connections; commercial package scope is determined separately. |
| OpenAI-compatible providers, including DeepSeek | Expansion through compatible APIs; compatibility does not mean enabled access for every customer. |
| Additional and self-hosted models | A development direction when need and total-cost advantage are established. |
Specific models enter a commercial offer after checks of quality, tool use, contractual terms and cost accounting. Technical support for a family does not mean all of its models are available in all plans.
Routing is already implemented as an engine component. Its economic impact needs validation on completed tasks with comparable quality. Switching models should not increase rework and repeat calls enough to erase the savings.
9. Data, trust and control
EXME processes tasks through the platform, the assistant workspace, model providers and connected services. The data transferred depends on the task and tool. Product policies should clearly explain processing purposes, retention, deletion and provider roles.
Managed infrastructure removes platform setup work. Users still authorize connections to email, calendars and other personal accounts. Access should match the request, and consequential actions should remain understandable and controlled.
Trust matters for retention: users need to see results, understand limitations and be able to stop work. Model errors require verifiable conclusions, limited authority and clear recovery from failure. These capabilities and their operation are part of service cost.
10. Product readiness and the next growth milestone
Key technical components have been implemented: assistant and environment management, the agent loop, memory, tool connections, the model registry, routing and expense tracking. They form the foundation of a managed service.
Commercial readiness depends on the complete user experience: a person connects, completes useful work, returns and understands the cost. The next milestone is to validate this journey in the initial segment and connect execution quality to retention and margin.
The next investment decision depends on pilot cohort results: revenue, retention, actual delivery cost and validated acquisition channels. The growth plan below sets out how to obtain that evidence.
11. Business model and monetization
Two planned price points: $25 and $50
Standard at $25 per month is the initial pricing hypothesis. Extended at $50 per month is planned for more intensive use. Package scope, included consumption and limits need pilot validation; these are planned terms, not a live public offer.
The subscription pays for access to a managed assistant and maintained workspace. Additional model and resource consumption may be charged separately with transparent limits. Team and specialized packages are potential extensions after demand is validated.
The central monetization measure is customer contribution after direct costs: AI, infrastructure, support, payments, refunds and partner payouts. Consumption accounting connects expenses to assistant activity; pricing decisions should use the full cost of an accepted result.
The interactive model compares $25 and $50. Its quick selector changes price only, so a higher-priced package needs separate checks of AI consumption, support, churn and CAC. A price increase does not establish unchanged demand.
12. Financial model and path to profitability
Economics of model routing
Route routine work to cost-efficient Chinese models such as GLM; use a stronger model when quality or task complexity requires it. The GLM family is part of the product’s technical foundation; DeepSeek is a compatible expansion option, without a claim of current customer availability. Routing savings remain a planning hypothesis until measured on completed tasks.
Effective AI cost = (1 − share) × strong + share × (economy + fallback × strong) + other AI costs. Shares are fractions in this formula. The two model budgets cover the same monthly workload per average paying customer. Request share approximates workload share only when average token volumes and input/output mix are comparable. Otherwise use a token-weighted workload share.
Growth-target assumption: a full workload costs $6 on the strong route or $2 on the economy route; route 75% to the latter, with no extra fallback or other AI costs assumed. The effective budget is $3: $3 lower per customer, or 50% of the comparison budget. This explains the existing $3 assumption, not an additional discount. Enter measured tool, voice and embedding costs separately, and account for retries. Quality failures can erase the saving.
Savings are already included in variable costs, operating profit and cash flow. The 24-month comparison holds revenue, customer growth and all other costs constant. It is a sensitivity analysis, not evidence of unchanged model quality or realized savings.
Provider pricing reference: Z.AI pricing. GLM-5.3-Flash: $0.15 input / $0.50 output per million tokens (1 Oct 2026). Monthly budgets above are separate assumptions.
The model starts with zero customers. Inputs include both prices, the premium share and its delivery-cost multiplier; first-month new customers, CAC and fixed costs and their monthly growth rates; churn, model routing, other AI costs, infrastructure, additional variable support, payment fees, refunds, partner payouts, startup costs and opening cash. Additional usage revenue is excluded.
For month t:
- Closing customers = opening customers × (1 − churn) + new customers.
- Average paying customers = (opening customers + closing customers) / 2 — a simplification assuming even movement during the month.
- Revenue = average paying customers × average monthly revenue per customer.
- Variable expenses = average paying customers × direct costs per customer + percentage expenses on revenue.
- Acquisition expenses = new paying customers × CAC.
- Operating result = revenue − variable expenses − acquisition − fixed expenses.
- Cash balance = previous balance + cash result − one-off expenses for the period.
Contribution per customer = average revenue − direct expenses − allocated percentage expenses. With positive contribution, the basic break-even threshold is fixed expenses divided by contribution, rounded up. The threshold with ongoing acquisition includes its monthly budget in the numerator. With nonpositive contribution, customer growth alone does not lead to profitability.
The interactive model calculates simple CAC payback as acquisition cost divided by positive monthly customer contribution. This excludes churn and fixed expenses. Scaling decisions should supplement this estimate with actual cumulative contribution from paying cohorts. No LTV is claimed before sufficient retention history is available.
The path to profitability: prove repeatable value → retain a paying cohort → measure and achieve positive contribution → find a reproducible acquisition channel → scale within available cash. Routing savings enter the base plan after measurement. Increased load requires recalculating infrastructure and support; their costs need not remain linear.
The conservative, base and growth-target scenarios differ in verifiable inputs, not just a desired profit date. The model shows operating results before tax and financing. An investment decision requires separate calculations of taxes, working capital and contractual obligations.
With a tariff mix, model and other AI budgets, infrastructure and variable support are multiplied by (1 − premium share + premium share × premium cost multiplier). The per-payment fixed fee is not multiplied; percentage fees apply to blended revenue. Mix and routing savings are each counted once.
Potential savings from crystallization
Baking repeatable chains into pipelines is an additional potential source of savings beyond choosing more economical models. A captured process may run with fewer LLM calls while retaining tool use and validation. The savings depend on task repeatability, exception frequency and maintenance costs.
These additional savings are not included in the scenarios below. After implementation, they should be measured on equivalent tasks using accepted outcomes, total calls and full delivery cost. Lower LLM spending does not mean zero execution cost.
Three planning scenarios
The three launch scenarios below use a single $25 price, zero initial customers, $30,000 opening cash and $5,000 startup costs. Other AI costs are initially zero and must be replaced by measured voice, tool and embedding costs. The separate growth case uses a $25/$50 mix, $150,000 opening cash, $20,000 startup costs and a non-zero other-AI budget. Cash amounts are modelling assumptions, not an investment ask or funds already raised.
| Assumption | Conservative | Base | Growth target |
|---|---|---|---|
| New paying customers / month | 50 | 100 | 220 |
| Churn / month | 8% | 5% | 3% |
| CAC | $40 | $25 | $12 |
| AI / customer / month | $9 | $6 | $3 |
| Infrastructure + support | $6 | $4 | $2 |
| Fixed expenses / month | $4,000 | $4,000 | $6,000 |
| First profitable month | Not reached within 24 months | 7 | 3 |
| Recovery of startup investment and losses | Not reached within 24 months | 14 | 6 |
| Total cash required | $84,508 | $23,018 | $14,605 |
Growth target at $25/month: 220 new paying customers per month, 3% monthly churn, $12 CAC, $3 AI, $1 infrastructure and $1 support per average paying customer; fixed costs $6,000/month. This requires validated low-cost acquisition channels, retention and measured delivery costs. These are targets, not achieved results or a return guarantee. To reach $3 in AI costs, validate model routing and context reuse on real tasks. To reach $12 CAC, test partnerships, referrals and organic channels with fully attributed costs; the model also reserves 3% of revenue for partner payouts. Do not increase spend until cohort retention and margin are measured.
All scenarios assume payment fees of 3.5% + $0.30, refunds of 2% of revenue and effective partner payouts of 3%. These are planning rates; they do not assert any bank's tariff or partner-program agreement. Fixed expenses are the budget for the team and shared services; it requires validation against actual quotes and does not mean all project labor has already been estimated.
In the base case, contribution per customer is $12.575; break-even requires 319 average paying customers without new acquisition, or 517 with monthly acquisition of 100 customers at $25 each. The required cash reserve of $23,018 includes the initial $5,000. With opening cash of $30,000, no additional financing is required in this scenario. The conservative case requires approximately $54,508 more.
Operating profit is calculated before taxes, financing and depreciation. The cash model assumes monthly payment, even customer movement and expenses paid in the same month; annual prepayments, receivables and taxes require a separate calculation. Change the inputs and inspect all 24 months.
Practical path to profitability
- First 30 days — measurement. Recruit 20–30 pilot users, test three scenarios, and record accepted outcomes, AI costs and support minutes. This is a recommended experiment, not existing statistics.
- Days 31–60 — a paying cohort. Test price and limits. Working experiment targets: at least 60% activation on day one, at least 40% returning to a useful task in week 4, and positive contribution after direct expenses. These thresholds are management hypotheses.
- Days 61–90 — channels. Test two channels with limited budgets, measuring CAC and retention separately. Increase spending only after two measured cohorts; target simple CAC payback within 3 months.
- After validation — scaling. For the base scenario, maintain 100 new paying customers per month and churn no higher than 5%, with an AI budget around $6 per customer and fixed expenses around $4,000. When results diverge, recalculate the model and adjust the package before increasing advertising.
A path to scale
Illustrative growth case: 65% on $25 and 35% on $50; ARPU $33.75. Starts with 300 new paying customers per month, growing 12% monthly. CAC starts at $25 and grows 2% monthly; churn is 3%. Premium delivery costs are 1.8× standard costs. The mix and churn are held constant across the customer base.
ARR is 12 × end-of-month MRR, not revenue earned during a year. Modelled revenue in year 2 is $6,256,495. Revenue in month 24 is $936,672, based on average customers; its operating profit is $504,828, before tax and financing.
| Month | Paying at month end | New paying | Acquisition budget | Revenue | Operating profit |
|---|---|---|---|---|---|
| 6 | 2,282 | 529 | $14,593 | $69,000 | $28,844 |
| 12 | 6,404 | 1,044 | $32,439 | $201,332 | $105,812 |
| 18 | 14,224 | 2,060 | $72,106 | $451,650 | $245,493 |
| 24 | 29,394 | 4,066 | $160,280 | $936,672 | $504,828 |
Operating setup: 1–2 people for support and owned servers. The scenario allocates $6,000/month to the team and other fixed costs; no additional per-customer support fee is assumed. Infrastructure starts at $1 per standard customer for power, connectivity, maintenance and a replacement reserve; premium usage raises this budget. This is an illustrative cost budget, not verified capacity.
The growth path depends on acquisition economics, retention and capacity validation. At this customer count, measure concurrent task load, server headroom, support requests and response times. Add expansion costs or reduce growth if current hardware and 1–2 support staff cannot meet service levels. Pipeline crystallization savings are not included.
Explore the growth case13. Go-to-market and repeatable sales
The first offer. Start with independent professionals who regularly need research and documents. Demonstrations should show a finished result on a recognizable task and shorten the path to a user’s first request.
Assisted launch. Run a limited pilot and observe setup and the first result. Hands-on team support helps identify friction but must be included in costs. The goal is to learn which tasks users will repeat without ongoing assistance.
A paying cohort. Test pricing, limits and reasons to renew. Compare activation, repeat tasks, retention, delivery costs and support by cohort. Set quality and positive-contribution criteria before expanding.
Channels. Test workflow demonstrations, professional communities, partnerships and referrals. Attribute all channel costs to acquired customers, including content, team effort and partner payouts. Organic acquisition is not automatically free.
Scaling. Increase spending where retention and payback have been established. Open new segments and team packages after validating their needs. Sales growth should be accompanied by control of quality, capacity and cash.
14. Roadmap and use of capital
Capital should reduce the main business uncertainties and fund progress to the next milestone. Financing needs and deal terms are determined by the next milestone’s budget and demonstrated results. The model’s $30,000 in initial funds is a cash assumption, not a claim of capital already raised.
Stage 1 — a useful product outcome. Spending priorities: engine reliability, onboarding, priority skills and integrations, observability and pilot support. Gate: selected tasks repeatedly end in outcomes users accept.
Stage 2 — execution economics. Spending priorities: full-cost measurement, routing quality, consumption control and pricing validation. Gate: positive contribution in a real paying cohort without deterioration in quality or retention.
Stage 3 — sales validation. Spending priorities: bounded acquisition experiments, partner scenarios and support for early cohorts. Gate: repeatable channel payback and clear cost attribution.
Chain crystallization into pipelines — a planned technology milestone. Once repeatable task quality is established, capture successful chains, introduce versions, validation, access controls and return exceptions to the agent. Gate: equally usable outcomes with less LLM computation and lower total delivery cost, including maintenance.
Stage 4 — expansion. Spending priorities: the plugin and skill catalog, team capabilities and infrastructure for validated demand. Self-hosted models are assessed on total cost. Gate: quality and contribution hold as usage grows.
Each stage’s budget should be refined using the previous stage’s results. Team, operations, security and a liquidity reserve must be accounted for before increasing marketing spending. Capital-allocation percentages require actual hiring plans and supplier quotes.
15. Risks and conditions for an investable business
Repeat use. Initial interest may not lead to subscription renewal. The main test is useful repeat work and paying-cohort retention; product focus should narrow to scenarios with durable value.
Execution cost. Complex work, retries and support can consume revenue. Responses include full-cost accounting, limits, quality-aware model selection and package revision before scaling.
Quality and trust. Incorrect conclusions and actions affect retention and reputation. Verifiable outcomes, limited authority, approval of consequential actions and failure handling are required.
Competition. Model providers and applications may expand their own agent features. EXME’s response is the quality of specific workflows, the convenience of a shared environment and disciplined service operation. This strategy’s effectiveness remains to be demonstrated.
Providers, data and licenses. Prices, APIs and contractual terms change; external tools expand data processing. Alternative connections, terms review, minimal access and compliance with component licenses are necessary.
Cash runway. Operating profitability is not the same as recovering startup costs. Acquisition spending, taxes and obligations may require additional financing. Growth decisions should consider cash flow and the conservative scenario.
16. Product status and forecast assumptions
As of this edition, EXME has an implemented technical foundation for a managed assistant. Chain crystallization, a broad ecosystem, commercial model packages, team capabilities and scaling are development directions. Hundreds of plugins and thousands of skills describe the catalog’s target scale.
Financial figures in this document are scenario-model outputs, not company financial statements. Price, new customers, churn, CAC and costs are testable assumptions. Profit timing and cash requirements follow from those inputs and change with them. Routing savings need validation against quality and actual costs.
The next investment decision should rest on accepted task outcomes, paying-cohort retention, full delivery cost, actual CAC and a cash budget. Commercial model and payment terms are checked with providers: Z.AI pricing, Anthropic pricing and Stripe pricing. These sources describe external terms; they do not establish EXME’s results.
Check the assumptions and all 24 months of calculations. The model supports discussion of the conditions for sustainable growth and how to validate them.