Pre-seed investor deck · August 2026

The memory layer for sovereign AI.

MemHoard.Ai and MemHoard.App are a completed local cognitive-memory product for SMEs and startups. The $5M round builds the company, formalizes its experienced engineering core and proves repeatable commercial adoption.

Financing targetUSD 5 million

Local · Sovereign · Persistent

Confidential · Forward-looking statements included

02 / 22

The investment thesis

AI has models. It has search. It still lacks a portable memory layer.

Linux made computing portable across heterogeneous machines. MemHoard.Ai aims to make persistent, governed memory portable across models, applications and hardware.

Commercial wedge nowCompleted private memory for SME files, databases and operations.

Expand horizontallyOne memory substrate for apps, agents, vehicles and robots.

Scale technicallyFrom a single node to routed experts and distributed compute.

Sources & framing

Analogy, not equivalence. The Linux comparison describes a strategic ambition for portability and standardisation; it is not a current market-share claim.

CNCF reports Kubernetes as a common operating layer for modern infrastructure, with 82% of surveyed container users running it in production in 2025: CNCF Annual Cloud Native Survey 2025. Android documents the Linux kernel as the platform base: Android Open Source Project.

03 / 22

The adoption gap

SMEs are being asked to adopt AI without owning the context it needs.

17%of small EU enterprises used at least one AI technology in 2025, versus 55.03% of large enterprises.

01

Knowledge is fragmented.

Documents, network shares, databases, images and voice notes are stored as separate assets—not as durable organizational context.

02

Retrieval is not continuity.

Search and RAG can locate content; they do not automatically maintain identity, chronology, relationships and operational state across time.

03

Cloud-first defaults create sovereignty and cost friction.

Regulated or IP-sensitive SMEs need data, models and memory inside their boundary—and predictable capacity for repeated agent workloads instead of open-ended token consumption.

Sources & definitions

AI adoption by size: 2025 EU survey, 17% small, 30.36% medium and 55.03% large enterprises: Eurostat, Use of artificial intelligence in enterprises (Dec. 2025). Survey scope is enterprises with at least 10 employees.

Distinction between external context, RAG and long-term agent memory: Letta context hierarchy. The State of FinOps 2026 documents growing concern around AI usage visibility, variable pricing, tokens and inference cost. Sovereignty requirement is MemHoard's product thesis, not a market statistic.

04 / 22

Competitive landscape · public documentation review

The gap is not search. The gap is persistent memory.

Solution category	Semantic retrieval	Cross-system connectors	Durable cross-session state	Entity / relationship memory	Multimodal operational memory	Source provenance	Fully local / air-gapped	Model agnostic	Shared with apps / machines	Federated node roadmap
OS / file search	Limited	No	No	No	No	File metadata	Yes	N/A	No	No
Microsoft 365 Copilot + connectors	Native	100+ / MCP	Product-scoped	Graph / index	Text-led	Citations	Federated source; cloud service	No	M365 ecosystem	No
Glean	Native	Native	Agent workflow	Enterprise graph	Content-led	Native	Not positioned as air-gapped	Platform models	APIs / agents	No
AnythingLLM	RAG / vectors	Extensions	Workspace history	Not a core graph	Document-led	RAG sources	Yes	Yes	API / agents	No
Agent-memory frameworks (Letta / Mem0)	Memory retrieval	Developer-built	Core capability	Structured memory	Implementation-dependent	Implementation-dependent	Self-host options	Yes	Agent-centric	Not the core product
MemHoard.Ai product / roadmap	Product	Selected sources	Product	Product	Product	Product	Default	Product	App + API	R&D roadmap

Native / documentedPartial, scoped or configuration-dependentNot a documented core capability

Sources, methodology & limitations

Method. Comparison reflects publicly documented competitor positioning reviewed 1 Aug. 2026. “Not a documented core capability” does not prove technical impossibility. The MemHoard row separates current first-party product claims from the federated-node R&D roadmap; live demonstration, connector inventory and independent benchmarking remain diligence requirements.

Microsoft semantic indexing and connectors: Microsoft Learn. Glean Knowledge Graph and agent workflow memory: Glean Knowledge Graph, Glean agent memory. AnythingLLM local/private positioning: AnythingLLM. Letta persistent memory blocks: Letta. Mem0 self-hosted memory: Mem0.

05 / 22

The product

Turn company information into a governed cognitive memory.

MemHoard.Ai is the completed, model-independent local platform that ingests authorized information, builds durable context and serves grounded memory. MemHoard.App is its application surface for people and operational workflows.

01

Remember relationships, not only chunks.

Connect people, files, places, decisions, procedures, time and provenance.

02

Keep memory under customer control.

Local-first deployment, role-aware access and explicit export or cloud choices.

03

Expose one memory to many surfaces.

MemHoard.App, APIs, enterprise agents, vehicles and robots.

MemHoard.Ai

INGESTFiles · DBs · media

UNDERSTANDEntities · context

REMEMBERTime · relations

ACTIVATEApps · agents · robots

Evidence status: MemHoard Ltd reports a complete product, testing across the stated local-hardware tiers and active commercial proposals. This public deck does not present named customers or booked revenue. First-close diligence therefore requires a live demo, reproducible benchmark pack and named paid-deployment evidence—no traction figure is inferred.

Sources & product status

Current product status and architecture are first-party statements by MemHoard Ltd, dated 1 Aug. 2026; the public product positioning is available at MemHoard.Ai. Demonstrations, security review, deployment records and benchmarks belong in the investor data room. The conceptual distinction between persistent agent memory and external retrieval is supported by Letta's context hierarchy and the Mem0 technical paper.

06 / 22

Commercial starting point · product complete

The product exists. The round builds the company around it.

MemHoard.Ai is local-first for sovereignty and privacy, with cloud only when explicitly requested. It is already available, tested and active in commercial proposals and PoCs.

01 · PRODUCT STATUS

Sellable now

The current product searches, reasons across and serves cognitive tasks on enterprise databases, network shares, file paths and documents.

02 · MEMORY FOUNDATION

Cognitive base

Documents, written and voice notes, images and diagrams become persistent cognitive concepts inside the customer-controlled MemHoard database.

03 · OPERATIONAL DEPTH

Beyond generic prompting

Responses can rely on path knowledge, database structures, chronology and extracted relationships—not only a single prompt or vector match.

04 · PILOT MECHANISM

Hardware-backed PoCs

Proposed model: lend a configured node for a multi-month KPI-based PoC. Failed KPIs return the hardware; passed KPIs activate a pre-agreed multi-year contract.

05 · GO-TO-MARKET

Commercial build-out

Build business development, marketing, solution selling, onboarding and customer success around the first production-ready MemHoard offer.

06 · COMPANY-BUILDING ROUND

$5M · 24 months

Formalize the six-engineer core, create the RAK operating organization, fund pilot hardware and prove a repeatable UAE/EU commercial engine.

Product, PoC model & evidence classification

Product completion, availability, testing, PoC activity and current operational scope are first-party MemHoard Ltd statements dated 1 Aug. 2026. Public positioning: MemHoard.Ai — The Memory Layer for Private AI. The loan-hardware/KPI/multi-year-contract mechanism is a proposed commercial structure, not a claim that such contracts are already signed. Exact PoC duration, KPIs, hardware liability, termination and contract terms require legal and customer validation.

07 / 22

Current sellable product · three workflows

Show the product through real operating patterns.

01 · CONTRACT MEMORY

“Find the signed Client X contract containing an exclusivity clause similar to Project Y.”

PDFsEmail threadsAttachmentsRelated projects

MEMHOARD.AI RESPONSE PATTERN

Documents + relationships + provenance

Found 4 relevant documents
Contract_2022_ClientX.pdf — illustrative 92% relevance
Related email thread — illustrative 87% relevance
Project Y final agreement — illustrative 84% relevance
Exclusivity clause extracted with source references

02 · PATH-AWARE FILE MEMORY

“Find the network file with the last approved packaging layout for Line 4. I only remember it was under the old 2023 procurement folders.”

File sharesFolder pathsFile namesArchive trees

MEMHOARD.AI RESPONSE PATTERN

Path logic + version context

Found 3 likely files
\procurement\2023\line4\layouts\pack_layout_v7.pdf
Matched folder history from legacy procurement path logic
Returned adjacent versions for rapid validation

03 · DATABASE REASONING

“Calculate maintenance time by machine for the last quarter and show which machines caused the highest downtime.”

Production DBMaintenance logsMachine IDsDowntime analysis

MEMHOARD.AI RESPONSE PATTERN

Schema understanding + grounded calculation

Identified relevant machine, event and maintenance tables
Built the query for total maintenance time by machine
Ranked downtime sources by machine and maintenance hours
Returned a readable summary with supporting figures

Product capability & claim status

The three workflows represent current first-party MemHoard product capabilities across documents, path-aware file discovery and structured databases. Example filenames, relevance scores, document counts and query outputs are illustrative response patterns—not customer benchmarks or measured accuracy claims. Production behavior remains permission-scoped and must be validated against each customer's repositories, database schema and security policy.

08 / 22

SME deployment wedge

Private AI now fits on compact, high-memory hardware.

01 · AMD AI MAX+ PRO

Ryzen AI Halo class

An x86 local-AI option combining CPU, Radeon graphics and NPU for protected SME workloads.

Up to 128 GB unified system memory

First NVIDIA DGX Spark in a two-node deployment

Second NVIDIA DGX Spark in a two-node deployment

02 · NVIDIA GB10

DGX Spark · single or dual

One local Grace Blackwell node, or distributed inference across two nodes over a direct 200 GbE ConnectX‑7 link.

128 GB single · 256 GB aggregate dual

03 · HIGH-MEMORY DESKTOP

Mac Studio · M3 Ultra

A single compact node for model families whose quantized footprint benefits from substantially more unified memory.

Up to 512 GB unified memory · Apple cites 600B+ LLMs in memory

128 GBAMD x86 local-AI tier

128 → 256 GBone or two GB10 nodes

512 GBone M3 Ultra high-memory node

Technical precision: the two-Spark configuration aggregates 256 GB across nodes; it is not one transparently pooled VRAM address space. Each Spark has two rear QSFP ConnectX‑7 connectors; NVIDIA's validated two-node topology uses one matching port per unit and NCCL/MPI for distributed work.

Official hardware sources

NVIDIA documents 128 GB unified memory, two rear QSFP ConnectX‑7 network connectors and model support up to 200B parameters—or 405B with two Sparks: DGX Spark Hardware Overview. Its official two-node guide specifies a QSFP/CX7 cable, a direct 200 GbE link and distributed workloads using NCCL/MPI: NVIDIA Spark Stacking and Connect Two Sparks.

Apple states Mac Studio with M3 Ultra can be configured to 512 GB unified memory and run LLMs over 600B parameters in memory: Apple Newsroom, 5 Mar. 2025; current specifications: Apple Mac Studio. As of the 1 Aug. 2026 research cutoff, Apple's current product page still identifies M3 Ultra; an M5 Ultra configuration is therefore not claimed. AMD documents up to 128 GB unified memory for Ryzen AI Halo class systems: AMD, 2026.

Model fit depends on precision, quantization, context length, KV cache, runtime and overhead. Manufacturer parameter figures are capability ceilings, not MemHoard latency or throughput benchmarks. MemHoard Ltd reports the 128/256/512 GB deployment bands as tested product configurations; detailed results belong in the data room.

09 / 22

Why SMEs need an owned AI option

Agent adoption turns token billing into variable operating cost.

98%

AI spend is now a FinOps scope

FinOps practitioners managing AI spend rose from 31% in 2024 to 98% in 2026.

$2.50 / $15

Input / output per 1M tokens

Current standard API price for GPT‑5.6 Terra; reasoning, tools and repeated context can add consumption.

$11k/mo

Transparent workload example

2B uncached input + 400M output tokens per month at the cited Terra rates.

For repeatable, high-duty workflows, MemHoard.Ai offers SMEs a second economic model: owned local capacity, private models and persistent memory—without mandatory per-token cloud dependency.

Sources, price basis & calculation

The State of FinOps 2026 covers 1,192 respondents representing more than $83B in annual cloud spend. It reports that 98% now manage AI spend (63% in 2025; 31% in 2024), AI cost management is the most desired skillset, and granular monitoring of tokens, LLM requests and GPU utilization is the leading requested tooling capability. The sample is the FinOps community, not all SMEs.

OpenAI's current model comparison lists GPT‑5.6 Terra at $2.50 per 1M input tokens and $15 per 1M output tokens: OpenAI API model comparison. Illustrative arithmetic: 2,000 × $2.50 + 400 × $15 = $11,000/month. This scenario assumes uncached text tokens and excludes tool calls, storage, networking and discounts.

Economic discipline. This is evidence of variable AI cost—not proof that local inference is always cheaper. MemHoard must benchmark total cost of ownership by workload, including hardware depreciation, energy, administration, model quality, latency and utilization.

10 / 22

Beachhead model · prove willingness to pay first

Do not sell to “all SMEs.” Win a specific operating profile.

ICPKnowledge-intensive SME · 50–249 staffPRIVATE DATA

BUYING TRIGGERRepeated AI/agent use + fragmented repositoriesOWNED CAPACITY

24-MONTH TARGET50 production customers × $40k blended ARR$2M ARR

Qualification before TAM

Repository complexity: High
Data / model sovereignty: Required
AI utilization: Recurring
Paid proof: Mandatory

The 2.04M UAE + EU organization universe is context—not a revenue forecast. MemHoard will publish a TAM/SAM only after paid deployments establish vertical mix, contract value and conversion.

Inputs, formula & limitations

UAE Ministry of Economy reported more than 557,000 SMEs and 63.5% of non-oil GDP in 2022: UAE Ministry of Economy & Tourism. Eurostat's 2025 enterprise-ICT survey population was ~1.53M enterprises with 10+ staff: 83% small, 14% medium, 3% large, yielding ≈1.484M SMEs: Eurostat metadata.

The former “organization count × assumed ACV” TAM has been removed from the main investment argument. The $2M exit-ARR target is forward-looking arithmetic: 50 production customers × $40k blended ARR. Both customer count and pricing require validation; the existing pricing hypothesis spans $18k–$60k annual platform revenue before services.

11 / 22

Business model · pricing hypotheses

Land with one memory node. Expand with usage.

01 · PILOT

Paid discovery + deployment

$15k–$40k hypothesis

One repository set, one measurable workflow and agreed success criteria.

02 · PLATFORM

Annual node / site license

$18k–$60k hypothesis

Software, updates, memory services, administration and support.

03 · EXPANSION

More domains and nodes

Recurring expansion

Departments, offices, databases, agents and higher-memory hardware.

04 · SERVICES

Integration + assurance

Project revenue

Data onboarding, custom connectors, governance, training and regulated deployment.

Hardware may be customer-owned, financed or bundled through partners. Pricing is deliberately presented as a hypothesis until pilot willingness-to-pay and support cost are measured.

Sources & assumption status

All prices, expansion mechanics and commercial packaging on this slide are management hypotheses for validation—not historical pricing, contracted revenue or third-party benchmarks. The revenue model should be updated after the first cohort of paid pilots.

12 / 22

Go-to-market

Sell measurable recovery of knowledge—not generic AI.

01 · TARGET

Document-heavy SMEs

Engineering, industrial, accounting, legal and technical-service firms.

02 · DIAGNOSE

Choose one costly memory gap

Lost contracts, drawings, procedures, decisions or maintenance context.

03 · PILOT

Private 8–12 week deployment

Pre-agreed corpus, users, security boundary and business KPIs.

04 · PROVE

Measure retrieval and adoption

Time-to-answer, answer traceability, successful tasks and active use.

05 · EXPAND

Convert and add nodes

More repositories, departments, applications and partner-led implementations.

Initial geography: UAE—using Ras Al Khaimah as the operating base—followed by selected EU markets where local control and knowledge continuity are strong buying drivers.

Sources & plan status

Target segments, pilot duration, KPIs, geographic sequence and channel strategy are management plans. They are not claimed historical results. UAE SME context is supported by the UAE Ministry of Economy & Tourism.

13 / 22

The model-size inflection

Kimi K3

A newly released open-weight MoE model illustrates why the next private-AI runtime cannot assume that every weight lives in one machine's memory.

2.8TTotal parameters

104BActivated parameters

1MContext tokens

“Local” must evolve from one box to one governed fabric.

≈1.4 TB

Nominal 4-bit raw weights

2.8T parameters × 0.5 byte, before scale metadata, runtime buffers and cache.

≈52 GB

Nominal active weights per token

104B activated parameters × 0.5 byte; routing still needs access to a much larger expert pool.

>512 GB

Beyond a top workstation tier

Dynamic loading, sharding or distributed experts become architectural questions—not optional optimization.

Primary model source & calculation

Moonshot AI publishes Kimi K3 as a 2.8T-parameter MoE with 104B activated parameters, 896 routed experts, MXFP4 weights and a 1,048,576-token context: MoonshotAI/Kimi-K3; technical report: Kimi K3: Open Frontier Intelligence (27 Jul. 2026).

Memory figures are transparent nominal arithmetic: parameter count × 4 bits ÷ 8. Actual deployment footprint is higher and runtime-specific. The slide does not claim MemHoard currently runs Kimi K3.

14 / 22

Scope discipline · one funded company, staged options

Win sovereign SME memory first. Earn the right to expand later.

0–24 MONTHS · FUNDED CORE

SME cognitive memory

Commercialize the completed local product, formalize the six-engineer core, prove customer ROI and build repeatable deployment, security and support.

Capital priority: customers, organization and product reliability.

POST-PROOF · PARTNER ADJACENCIES

Hospitality + mobility

Apply permissioned memory to robots and vehicles only through funded partners after the SME platform demonstrates retention and integration economics.

Gate: core revenue plus a partner-funded deployment.

LONG-TERM · RESEARCH OPTION

Distributed model fabric

Dynamic loading, routed experts, WAN compute and eIDAS-aligned settlement remain strategic research options—not the 24-month commercial plan.

Gate: separate technical benchmarks, capital plan and regulatory case.

Technical precedents & caveats

The 24-month funded scope is a management decision introduced to separate commercial execution from future options. Hardware and token economics appear in the optional technical appendix. Hospitality, mobility, Kimi-scale deployment, dynamic offload and distributed settlement remain documented there for diligence without being represented as near-term revenue commitments.

External technical precedents include DeepSpeed ZeRO-Inference, NVIDIA Megatron Core and Petals. They support feasibility research, not a claim of current MemHoard implementation.

15 / 22

Distributed capacity + verifiable settlement

A network needs more than routing. It needs evidence and incentives.

DISCOVERNodes advertise permitted models, experts, memory and compute.

Policy decides which workloads may leave a customer boundary.

EXECUTEPrompts or model states route to eligible capacity.

Work is metered; sensitive payloads require encryption and explicit controls.

ATTESTRequests, outputs and usage records are cryptographically signed.

A permissioned ledger can preserve tamper-evident ordering and settlement evidence.

REWARDVerified providers receive usage-based compensation.

Commercial, regulatory and token-design questions remain open R&D.

Legal precision: a blockchain record is not automatically a notarized or qualified eIDAS signature. Qualified legal effects require the applicable eIDAS controls and qualified trust-service providers.

MemHoardPOLICY · ROUTING · MEMORY

SME nodeLocal model + memory

Expert nodeRouted MoE capacity

Compute nodeCPU / accelerator

Trust nodeAttestation + ledger

eIDAS, ledger sources & company claim

Under eIDAS, qualified electronic signatures have the equivalent legal effect of handwritten signatures: European Commission eIDAS overview. Regulation (EU) 2024/1183 states that electronic ledgers cannot be denied admissibility solely because they are electronic, while qualified ledgers receive presumptions regarding ordering and integrity: EUR-Lex.

EBSI demonstrates permissioned blockchain trust registers and verifiable credentials: EBSI Verifiable Credentials. Management states that current MemHoard units include validator-node functionality; this deck does not independently verify production status, conformity assessment or qualified trust-service status.

16 / 22

Embodied memory use case · Ras Al Khaimah hospitality

A concierge robot should remember the guest journey.

1.35Movernight visitors to Ras Al Khaimah in 2025, up 6% year-on-year; tourism revenue grew 12%.

01 · CHECK-IN

Create a consented stay memory

Guest identity token, language, room, preferences, authorized companions and service requests.

No facial template without explicit legal basis and consent controls.

02 · DURING STAY

Recognize context, not just a face

“Welcome back” is grounded in active-stay status, location, role and permitted history.

Least-privilege access; local processing where possible.

03 · SERVICE

Carry commitments forward

Remember that towels were requested, a restaurant was booked or accessibility support is required.

Every action retains source, time and staff override.

04 · CHECK-OUT

Respect the end of the stay

Detect checkout timing, close operational tasks and apply retention/deletion policy.

Memory lifecycle must follow hospitality and privacy policy.

Regional timing: RAKTDA targets more than 3.5M visitors by 2030. Wynn Al Marjan Island is expected to open in 2027 with 1,500+ rooms and more than 20 restaurants and lounges.

RAK, Wynn, robotics & privacy sources

RAK tourism: 1.35M overnight visitors in 2025, +6% year-on-year, +12% tourism revenue, and ambition to exceed 3.5M visitors by 2030: RAKTDA, 19 Jan. 2026. Wynn scope and expected 2027 opening: Wynn Resorts 2025 Form 10-K.

Hospitality-robot acceptance research (358 UK hotel guests) finds trust among important adoption factors: International Journal of Hospitality Management, 2025. Under GDPR, biometric data used to uniquely identify a person is sensitive data: European Commission. The workflow is a future MemHoard concept, not a deployed hotel product.

17 / 22

Personal vehicle + robotaxi use case

“Go pick up my wife at work.”

A navigation system can parse an address. A cognitive-memory layer must resolve authorized identity, the correct workplace, entrance, timing, passenger verification and exceptions—without exposing the household memory to every vehicle subsystem.

The hard problem is context orchestration.

WHOHousehold identity

Resolve “my wife” through explicit relationships and permissions.

WHEREPlace memory

Workplace, correct entrance, pickup zone and learned geography.

VERIFYPassenger match

Consent-based face, device or credential verification.

ACTSafe hand-off

Arrival notice, exception handling and auditable completion.

The same memory primitive can serve owned vehicles and autonomous-taxi fleets, with different tenancy, consent and retention rules.

Autonomous-mobility context & privacy

Dubai RTA's strategy targets 25% of journeys via autonomous transport by 2030. RTA reported pilot operations by Baidu Apollo Go, WeRide and Pony.ai and a planned commercial driverless launch in 2026: Dubai RTA. The “pick up my wife” sequence is a MemHoard future-use-case hypothesis, not a claim about a current vehicle integration. Biometric controls reference European Commission GDPR guidance.

18 / 22

Founder + six-engineer core + scientific guidance

An experienced technical nucleus—not a team to invent after funding.

Founder · Product vision

Matteo Compagnoni

Created MemHoard after seeing industrial project knowledge disappear when contracts closed and teams changed.

25 years in multinational engineering and R&D
Company-reported portfolio of six international patents
IoT, sensor-edge and data-platform experience

Historical engineering core

6 senior engineers

Six experienced engineers who have worked in the sector for years form the technical competency base behind the company-building plan.

Senior execution rather than a graduate-only hiring thesis
Round formalizes contracts, ownership, roles and delivery cadence
Names, commitments and responsibilities disclosed in diligence

Professor Angelo Cangelosi with the iCub robot

Scientific advisor

Prof. Angelo Cangelosi

Professor of Machine Learning and Robotics at the University of Manchester; founder and co-director of its Centre for Robotics and AI.

Cognitive and developmental robotics
Language grounding and human–robot interaction
Long-term research guidance, not operating headcount

Team sources & evidence status

Founder experience, patents and the historical six-engineer core are first-party MemHoard Ltd statements. Before closing, the data room should identify all six engineers, relevant experience, intended roles, contractual status, time commitment, compensation and option allocation. Founder profile: Matteo Compagnoni. Angelo Cangelosi's academic roles: University of Manchester; Developmental Robotics, MIT Press. The advisor relationship remains a company representation.

19 / 22

What the round actually builds

Convert a technical core into a durable operating company.

OPERATING COMPANY

MemHoard Ltd becomes the staffed Ras Al Khaimah organization: employment, governance, finance, legal, IP ownership and partner contracts.

PRODUCT + ENGINEERING

MemHoard.Ai formalizes the six-engineer core into accountable platform, data, security, QA and deployment ownership.

COMMERCIAL DELIVERY

MemHoard.App anchors sales engineering, onboarding, customer success, support and measurable business outcomes.

MemHoard Ltd is based at Innovation City Business Centre, RAK BANK ROC Office, Ground Floor, Al Rifaa, Sheikh Mohammed Bin Zayed Road, Ras Al Khaimah, United Arab Emirates.

Company location, organization & naming

Innovation City describes itself as Ras Al Khaimah's technology-focused free zone and publishes the cited address: Innovation City — About. Legal-entity status, the six-engineer formalization plan and organizational design are first-party statements and should be supported by the company license, IP assignments, employment/contractor agreements, governance chart and hiring plan in diligence. Brand roles are management-defined.

20 / 22

The ask · company-building round · 24 months

$5M

Build the organization around the completed product and its six-engineer core—then prove a repeatable SME commercial engine.

45%People: formalize the core, key hires and employee/option structure$2.25M

20%Product reliability, connectors, security, QA and deployment tooling$1.00M

15%UAE/EU sales engineering, customer success and channels$0.75M

10%Local AI hardware, lab, assurance and compliance$0.50M

10%RAK operations, legal, finance, IP and contingency$0.50M

Allocation, runway & organizational assumptions

Raise size and allocation are management proposals. Percentages sum to 100% and dollar values to $5.0M. Straight-line gross budget capacity is approximately $208k per month over 24 months before revenue; actual burn must be lower initially and milestone-gated. The operating plan targets formalization of the founder + six-engineer nucleus, 10–12 FTE by month 12 and 14–16 FTE by month 24, subject to hiring and commercial evidence.

The data room should include a monthly cash model, compensation bands, employment and option structure, hiring sequence, hardware capex, revenue assumptions and a minimum-cash trigger. This deck does not represent the allocation as an audited budget.

21 / 22

Milestone-gated organization + commercial proof

Release capital against evidence—not roadmap ambition.

0–6 MONTHS

Company formation

Formalize six-engineer roles and IP, establish governance and finance, publish benchmark/security pack, secure 3 paid lighthouse deployments.

6–12 MONTHS

Commercial validation

Target 10 paid deployments, 6 production conversions, referenceable ROI and validated pricing, support load and sales cycle.

12–18 MONTHS

Repeatable delivery

Target 25 production customers, channel-ready implementation, customer-success playbook and net retention baseline.

18–24 MONTHS

Series A evidence

Target 50 production customers and $2M+ exit ARR; future-use-case R&D proceeds only through partner funding or separate gates.

Product riskGate claims with third-party benchmarks and customer references.

Sales-cycle riskUse paid, narrow pilots with pre-agreed conversion criteria.

Team riskDocument names, commitments, IP assignment and ownership before first close.

Scope riskKeep robotics, mobility and distributed compute outside the funded core until commercial gates pass.

Target status & calculation

All milestones, customer counts, headcount and ARR figures are forward-looking targets—not forecasts. $2M exit ARR across 50 production customers implies $40k blended ARR per customer, within the current $18k–$60k platform pricing hypothesis but not yet validated. Named customers, signed contracts and measured revenue have not been supplied for inclusion; no traction has been invented.

A staged financing structure may tie portions of the round to company-formation, paid-deployment, conversion and ARR gates. Exact tranche terms are subject to investor negotiation and are not proposed as binding terms in this deck.

22 / 22

The memory layer for sovereign SME AI

A completed product. An experienced core. A company ready to be built for scale.

The $5M round formalizes the six-engineer nucleus, creates the operating organization and targets 50 production customers and $2M+ exit ARR in 24 months. Robotics and distributed compute remain gated future options.

Request investor diligence Explore the MemHoard website

Source methodology & forward-looking notice

Research cutoff: 1 Aug. 2026. External facts and technical precedents are linked on the slide where used. Priority was given to government, regulator, company documentation and primary research sources.

Deck structure: the principal investor narrative is 17 slides, including the completed SME offer, three current product workflows and local-hardware deployment evidence. Five deeper technical and future-use-case slides are hidden by default and available through the appendix button.

Evidence labels: “management representation” identifies first-party MemHoard information not independently verified; “target,” “hypothesis,” “roadmap” and “illustrative” identify forward-looking content. Named customers, signed revenue and measured performance are not inferred where evidence was not supplied.

No affiliation: Apple, AMD, Intel, NVIDIA, Kimi/Moonshot AI, Microsoft, Glean, AnythingLLM, Letta, Mem0, Wynn Resorts and other third parties are referenced for context only. MemHoard does not claim affiliation or endorsement.

MemHoard Ltd · Innovation City · Ras Al Khaimah · UAE