archie
Your AI colleague you actually own.
Executive Summary · Confidential · 2026
archie chunk is your physical AI colleague - a precision-machined device that sits on your desk, connected to your laptop by a short cable. Plug it in and he starts working: running the skills that eat your day, watching your screen for what he could take off your plate, learning how you work specifically, and keeping score of every hour he saves you - always looking for how he can ease your work even more.
AI is the most powerful productivity tool in a generation. But every tool available today comes with the same hidden costs: your data leaves your machine, your workflows are stored on someone else's servers, and your costs scale with every prompt you send.
Every AI tool you use today processes your data on servers you don't own, under policies you didn't write, in jurisdictions that may not align with yours. For lawyers, doctors, and financial advisers, this is not a privacy concern - it is a professional liability. For everyone else, it is a dependency that grows with every tool you add. And as Apple, Microsoft and Google integrate AI ever deeper into the operating system itself, platform-level control lives in a settings toggle - and toggles can disappear. archie chunk runs entirely on its own dedicated processor with isolated memory. Your laptop provides power and a screen. Nothing more. That is autonomy: not a promise you have to trust, but a switch you hold.
Privacy is physics. Not plugged in - not reachable, not runnable, not there.
Frontier AI token costs for professional teams run into several thousand dollars per month - variable, hard to budget, and rising with every new use case. Most of that spend covers tasks a smaller local model handles just as well. archie chunk eliminates token costs for the vast majority of tasks that follow a pattern. Cloud access remains available for the tasks that genuinely need frontier-model reasoning - on your terms, at a fraction of the cost.
Your work. Your machine. Your rules.
Your professional identity - on the chunk
archie learns you over time. Your writing style, your tone, your decisions, your context - how your firm handles certain situations, what worked before, what didn't. This professional identity is stored on the chunk. It travels with you. And it belongs entirely to you - not to a cloud platform, not to an employer, not to archie.
archie - Standard, Copper and Titanium editions
archie chunk runs the model on its own dedicated AI processor - completely isolated from the host. Just plug in the chunk, connect the cable, and archie starts. The laptop or Windows PC provides only power and I/O. No host RAM is used, no host CPU is involved in inference. No internet required, no tokens consumed, no data leaving the device. For tasks that genuinely benefit from a frontier model, archie optionally reaches the Anthropic API - processed but never stored, covered by a Zero Retention DPA. You decide which tasks go where.
Runs on the chunk's dedicated AI processor with isolated RAM. No internet needed. No tokens, no API costs. The host OS cannot access what runs here. All sensitive data stays on the device.
Anthropic API with Zero Retention DPA. Processed, never stored. For complex reasoning tasks. User controls what goes here. Non-sensitive content only.
"90% never leaves your laptop. The other 10% never gets stored."
Most tasks run entirely on-device. No per-query API costs for daily use.
Access to frontier models on demand - without compromising on what stays local.
Lawyers, doctors, tax advisers - sensitive mandates stay strictly local, always.
For professions handling sensitive client information - medicine, law, therapy, finance, tax, HR, consulting - cloud AI is not just uncomfortable. It is legally problematic. The 90% local architecture is not a feature - it is the foundation. Sensitive client data, mandates, and patient records never leave the local model. DSGVO compliance and relevant ISO certifications are achievable on this basis.
| Security Feature | What it means in practice |
|---|---|
| No persistent state | archie keeps your session strictly isolated. Unplug archie and credentials, state, context are taken with it. |
| Physical Kill Switch | The off switch is physical. Remove the chunk and the session ends in under 500ms - cleanly, completely, by design. |
| Isolated local processing | The AI model runs on archie chunk's dedicated processor with its own RAM - physically separate from the host. The host OS has zero access to the inference process. No internet, no API costs, no data leaving the device. Works identically on laptop and Windows. |
| Runs only when plugged in | Present when plugged in. Gone when not. No background activity, no idle listeners, no scheduled tasks. |
| PIN-protected hardware | Your archie chunk is protected at the hardware level by a PIN. A chunk without its PIN is exactly what it looks like: a beautiful piece of metal. |
| Zero Retention Cloud | For tasks requiring frontier-model reasoning, archie reaches the Anthropic API. Covered by a Zero Retention DPA: processed but never stored, never used for model training. Default: local only. |
| Your identity stays with you | archie stores your professional identity on the chunk - writing style, context, workflows, institutional knowledge. No cloud platform holds it, no employer can access it. |
| No training on user data | Local model: runs offline, no telemetry. Anthropic API with Zero Retention: contractually excluded from model training. |
| § 203 StGB / GDPR Art. 9 | archie Local meets the technical and organisational measures required for § 203 StGB and GDPR Article 9. Verification pending. |
There is a well-established pattern in technology: physical objects create relationships that software cannot. The Toniebox streams audio any phone could play - but children bond with the box. The Kindle could be an app - but readers choose the device. The Leica could be replaced by a smartphone - but photographers buy the Leica. archie works the same way: it sits on your desk, develops a patina, and becomes uniquely yours as it learns your workflows, your language, your exceptions.
You do not forget something sitting on your desk. You do not cancel something you reach for every morning. You do not replace something that has become personal. A colleague who has worked with you for a year is not interchangeable. Neither is archie.
The most desirable objects in the world don't just serve a function - they carry an identity. Rimowa collaborates with Supreme. Montblanc with BMW. Lamborghini with almost everyone who wants to signal that they take craftsmanship seriously. archie chunk is the first AI product that can do the same. A precision-machined object that sits on your desk every day, in a material you chose, with a finish that can carry a brand. Limited co-editions with the right partners don't just generate revenue - they place archie in a different conversation entirely: not "AI tool" but "object worth owning." No software product can do this. An app doesn't get a Rimowa collab.
"CEOs are frustrated by skyrocketing AI token costs, fueling interest in pivoting away from the closed models."
- Alex Karp, CEO Palantir
The largest professional services firms - consultancies, law firms, auditors, banks - are moving toward local AI. Regulatory pressure, client confidentiality, and token costs at scale make cloud-first AI increasingly untenable for sensitive work.
But local AI on a laptop has a structural problem: RAM. Running a capable model locally competes directly with the applications running alongside it. The solution is a dedicated device that handles AI inference independently, without touching the host machine.
archie chunk is that device. Deployed as a second device alongside the firm laptop, pre-loaded with firm-specific knowledge and personalised to each user. Branded to the firm. Owned by the employee. (And unlike cloud AI subscriptions: switch from Opex to Capex, depreciable immediately.)
archie titanium
archie aluminium
archie chunk is not just the delivery mechanism for the software - it is the product's core strategic asset: a hardware kill switch no software can replicate, margin that funds its own distribution, and a physical object that becomes irreplaceable as it learns you.
A hardware kill switch no software can replicate. Unplug = everything gone in <500ms. A policy cannot do this. Physics can.
Hardware margin ($241 at scale) covers the full $160 CAC — every unit generates +$81 net from day one.
People form relationships with physical objects in a way they never do with software. Churn drops structurally.
A software clone can be built in weeks. A physical product with brand and trust history cannot be copied overnight.
Precision-machined. Every detail deliberate.
| Edition | Material | RAM | Price |
|---|---|---|---|
| Standard | Aluminium | 8GB | $599 |
| Copper | Copper C101 OFE | 16GB | $649 |
| Titanium | Titanium Grade 5 | 16GB | $699 |
| Edition | Material | Price | Unit Cost (1k+) |
|---|---|---|---|
| Gold Ltd. | Solid Gold · 100 pcs | $15.999 | ~€9.500 |
macOS 13+ and Windows 11. The model lives on the chunk, runs on its dedicated AI processor. When local isn't enough - optional cloud access, Zero Retention, always your call.
Integrations:
Every request starts local. Only if a task genuinely needs more reasoning power - and contains no protected data - does it escalate to the cloud. You see and control every step.
archie's skill library will grow to hundreds of workflows. Two profiles anchor the strategy - everything else expands from there.
The only AI assistant that eliminates the conflict between productivity and data protection. With archie Local, client data never leaves your device.
CEO, CFO, assistant and accountant in one person. archie is the first employee you can afford before you send your first invoice.
Work freely on sensitive client and project data without exposing it. archie processes locally, acts locally, leaves no user data on the host.
You measure everything except the 8 hours you work. archie brings the same optimisation mindset to your professional life.
Children get AI assistance without their homework, questions, or personal data stored in a cloud profile. Parents control what skills are installed.
archie chunk is cashflow-positive from the first unit sold (+$109 Titanium / +$138 weighted after CAC). At 399k paid subscribers, the model reaches $97.5M ARR, $96.1M EBIT (41% margin) and $131.8M cumulative cashflow by Year 3 - with zero external capital required.
| Stream | Net Price | Net Margin (1k+) |
|---|---|---|
| Hardware - Standard (Alu, 8GB) | $503 net one-time | ~56% ($283 GM) |
| Hardware - Copper (16GB, Axelera) | $545 net one-time | ~33% ($180 GM) |
| Hardware - Titanium (16GB, Axelera) | $587 net one-time | ~36% ($209 GM) |
| Gold Ltd. Edition | $13.445 one-time | ~98% |
| Subscription Basic | $25 / month | $15 CM |
| Subscription Pro | $39 / month | $14 CM |
| Weighted CM (30% free / 50% Basic / 20% Pro) | - | $10,30 / month |
| Unit Economics (Titanium, net) | 100 units | 1.000+ units |
|---|---|---|
| Hardware net price | $587 | $587 |
| BOM + assembly + packaging (16GB Axelera) | $478 | $378 |
| Hardware GM | $109 (~19%) | $209 (~36%) |
| Hardware GM after CAC ($100) | $9 | +$109 from day one |
| Sub Revenue 12mo (weighted $20,30/mo) | $244 | $244 |
| Sub CM 12mo (weighted $10,30/mo) | $124 | $124 |
| LTV Revenue 12mo | $831 | $831 |
| LTV Revenue 24mo | $1.075 | $1.075 |
| LTV CM 12mo | $118 | $333 |
| LTV CM 24mo | $242 | $457 |
| LTV:CAC Revenue 12mo | 8.3:1 | |
| LTV:CAC Revenue 24mo | 10.8:1 | |
| LTV:CAC CM 12mo | 3.3:1 | |
| LTV:CAC CM 24mo | 4.6:1 · industry minimum: 3:1 (CM basis) | |
| Revenue basis: total customer revenue over period. CM basis: after hardware BOM and subscription costs - the conservative, investor-grade metric. Weighted mix (50% Alu / 30% Copper / 20% Titanium): ASP ~$527 net, BOM ~$289 (Alu: $220 / Copper/Ti: ~$370 avg), GM ~$238 (~45%), +$138 after CAC ($100). | ||
At 1.000+ units (Titanium): hardware GM $209 covers full CAC $100 → +$109 from day one. Weighted mix GM $238 → +$138 after CAC. Subscription compounds on top: +$124 CM in year one alone.
archie sells itself when seen. The entire strategy is built around creating the moment of seeing - in the right communities, at the right time, with the right people.
No hardware. Landing page + 90-second demo video. Personal network, Product Hunt, Hacker News, Reddit.
First production run - copper and aluminium. Deposits and hardware revenue fund the bulk of production.
Full range across all editions. Paid acquisition funded entirely by hardware margin. B2B becomes primary revenue driver.
International expansion - UK, US, Canada, Australia. Subscription ARR compounds. Exit conversations become natural.
| All figures in USD | Launch 1.000 units | Year 1 21.000 units | Year 2 171.000 units | Year 3 571.000 units |
|---|---|---|---|---|
| New units sold | 1.000 | 20.000 | 150.000 | 400.000 |
| Paid subscribers | 700 | 14.700 | 119.700 | 399.700 |
| Hardware Revenue (wtd. $527) | $527k | $10.5M | $79.3M | $211M |
| Hardware CM (~45%) | $238k | $4.76M | $35.8M | $95.5M |
| ARR (subscription) | $171k | $3.6M | $29.2M | $97.5M |
| Subscription CM | $21k | $900k | $14.7M | $49.3M |
| Gross Profit | $159k | $3.66M | $35.4M | $104.5M |
| Fixkosten (Team, Mktg) | -$56k | -$476k | -$3.0M | -$8.4M |
| EBIT | +$103k | +$3.18M | +$32.4M | +$96.1M |
| EBIT-Marge | ~20% | ~27% | ~37% | ~41% |
| Cumulative Cashflow | +$103k | +$3.3M | +$35.7M | +$131.8M |
70% of users opt into a paid subscription ($25 or $39/month). 30% use archie free via the local model — and still represent future upgrade potential. Maximum capital requirement is ~$500k (Step 1-2, higher BOM vs. original version), covered by technology grant.
archie sits at the intersection of three categories attracting significant capital: AI tooling, privacy infrastructure, and hardware-software consumer products.
The AI supercycle window - where AI companies command 15-25x revenue - will not last indefinitely. archie reaches $29.2M ARR by Year 2 (EBIT $32.4M), squarely within this window.
No external capital, no pressure. Hardware margin funds growth, subscription ARR compounds to $97.5M ARR and $131M+ cumulative cashflow before any exit conversation begins conversation begins.
AI model providers are seeking mass-market distribution. archie is a trusted, physical on-ramp to AI for users who will never set up a CLI tool.
No competitor has a comparable physical product.
~45% weighted hardware gross margin ($238 absolute per unit at 1k+) - covers full $100 CAC with +$138 left over. Titanium edition: $209 GM, +$109 after CAC.
No external capital needed to reach profitability.
The 90% local architecture is auditable - no competitor can say the same.
archie chunk requires a USB-C port and PCIe bandwidth. Works on laptop and Windows, but not on devices without USB-C.
A lost or stolen chunk means lost workflows. New supply chain competencies with real capital risk.
Skills break when websites change.
Privacy-conscious laptop users who need archie but can set it up.
Millions arriving who want the benefit without the complexity.
One firm buys 50 sticks - distribution built on professional trust.
User-generated skills increase value without archie building every workflow.
Would triple the addressable market.
Apple and Microsoft are moving toward local AI at the OS level. The physical form factor buys time.
Core product depends on the Claude API.
LinkedIn already blocks Playwright. Others will follow.
A single credible security claim could trigger a trust collapse.
Each element exists in isolation. The combination is what makes this unique: market timing, category pioneer, self-funding model, and broad exit optionality.
The AI market is experiencing historic investment volumes. As agents penetrate daily life, the demand for autonomy and control grows - and intensifies as AI begins to see everything on the laptop through the operating system itself. Physical objects create that feeling in a way software never can. archie gives that control back: a physical object that runs when you plug it in, stops when you pull it out.
The market is splitting: cloud-first tools that do everything automatically, and physical products that put the user in control (hardware wallets, camera covers, Toniebox, Plaud). archie is the first to combine both - a software agent with the depth of the best AI tools, delivered through a physical object that makes autonomy and control tangible.
Hardware margin funds growth, CAC is covered and internationally scalable. A software clone is buildable in weeks - a physical product with brand, daily desk presence, patina and trust history commands a moat that grows over time.
Three structurally different exit paths: enterprise software acquirers, AI model providers seeking physical mainstream distribution, hardware brands with subscription ambitions. Plus the IPO window of the AI supercycle. No single buyer needs to say yes.
Because software can't be unplugged. archie chunk runs when connected. The model runs on the chunk's own dedicated AI processor with isolated RAM - the host laptop or Windows PC has no access to what happens inside. Unplug - and the session ends. Completely. No process keeps running, no data lingers, no cloud keeps a copy. The host OS sees a USB device. It cannot inspect, log, or interfere with what runs on the chunk. That guarantee is physically impossible for a pure software product to make - regardless of its privacy policy.
75,000 is the cumulative total over 24 months - ~3,125 units per month. For context: Oura Ring sold ~150,000 units in their third year at a similar price point. reMarkable reached ~100,000 in Year 3. Superhuman had ~75,000 paying users after three years - with no hardware.
Three compounding drivers: B2C self-optimizers, compliance-driven professionals who cannot legally use cloud AI with client data, and B2B corporate - a single company deploying 50 units covers 0.07% of the target.
Tiiny AI Pocket Lab raised $3M on Kickstarter — proof that the market for local AI hardware is real. But it's a 300g, 65W device at $1.399, aimed at developers who want to run 120B models locally. archie chunk is the opposite answer: 80g, 6.5W, $499-599, for professionals who want control, security and productivity — not a local AI server on their desk. Two completely different customers.
Littlebird raised $11M in March 2026. laptop + Windows, always-on screen context, 90+ integrations - the closest in philosophy, but explicitly cloud-synced. Their killer feature (context available everywhere) is exactly what makes them incompatible with regulated professions.
archie and Littlebird are not competing for the same customer. Littlebird's user wants seamless cloud memory. archie's user wants an AI colleague that exists only when they want it, leaves no residue, and can handle their most sensitive data.
The deposit model means production is only triggered after payment. Hardware gross margin at 100 units (Copper, net) is ~54% - each unit sold covers its own CAC before a single subscription is collected.
The $200-300k bridge needed in Step 2 is grant-eligible - technology innovation grants at regional and national level cover exactly this stage. Comparable precedent: reMarkable funded their first production run entirely through pre-orders.
Being European is a structural advantage in this specific category. US companies must retrofit DSGVO compliance as an afterthought. archie was designed around it from day one.
The track record is strong: Skype, Spotify, ElevenLabs, Mistral, Aleph Alpha, Celonis. What matters is insight, execution and timing - none of which are geography-dependent.