Maxy.
April 2026
Knowledge workers and solopreneurs are drowning in admin.
And the “solutions” make it worse.
The reality for the individual contributor
- Output is a function of their own thinking — every hour on admin is an hour off the work that pays
- Relationship signal lives across email, calendar, WhatsApp, Signal, Telegram, LinkedIn — no single tool sees all of it
- Context dies between sessions — every conversation re-establishes who, when, why
- One person cannot scale their own attention without losing the thinking that makes the work valuable
Why existing tools fail
- CRMs require data entry — the thing the IC does not have time for
- Automation tools translate English into flowcharts — a new paradigm to learn
- Every “solution” means another dashboard, another login, another app
- AI tools promise full automation but cannot deliver it reliably — and when they fail, the human has forgotten how
The enemy is not AI that generates noise.
It is AI that captures, then hands off to someone else's cloud.
| Everyone else says | Maxy says |
|---|---|
| “AI replaces the task” | “AI earns the right to the task” |
| “Set it and forget it” | “Trust, then verify, then trust more” |
| “You never have to do this again” | “You will always be able to do this. You will not have to.” |
| “Push capture to HubSpot, Slack, Salesforce” | “Route capture into one operator-owned graph, on your device” |
| “Cloud-powered, always available” | “Your data stays local. Your knowledge stays yours. When the cloud falls, you do not start from zero.” |
Four architectural axes: on-device, graph-first, schema-bounded LLM judgement, packaged for non-technical operators. Each one expanded on The Moat.
03Two independent signals legitimised the category
in April 2026.
GBrain — Garry Tan, 9 April 2026
- CEO of Y Combinator open-sourced his personal AI memory system. MIT licensed. 4,800+ stars and 541 forks in 24 hours.
- Tan's own archive: 10,000+ Markdown files, 3,000+ people pages, 280+ meeting transcripts, 13 years of calendar data, 40+ skills, 20+ cron jobs.
- Architecture: Markdown brain repo, Postgres with pgvector, agent-skills layer over OpenClaw and Hermes. Production-grade. Developer-facing.
LLM-Wiki — Andrej Karpathy, same month
- Published a parallel Markdown-first personal-memory pattern in a public gist.
- Two of the most credible technical figures in the industry, independently, inside a single month. The category is real.
- Both require the operator to also be a developer. Neither is packaged. Neither is graph-first. Neither is on-device by default.
One thesis. Four proofs.
Your business, understood
Structured graph memory — customers, jobs, invoices, suppliers, communications. Local embeddings and standards-based data (Schema.org, vCard, iCal, UBL). Every conversation starts with full business context. The intelligence accumulates and never leaves.
Any process, any complexity, plain English
Describe a process in conversation — simple or branching, one-off or recurring. Maxy orchestrates it using the right combination of tools, specialists, triggers, and scheduled events. Critical steps are guaranteed. Timing is automated. Specialist agents handle domain isolation. Natural language is the runtime, not the input.
Your relationships, surfaced on demand
Quantified strength across every channel the IC actually uses — email, calendar, WhatsApp, Signal, Telegram, LinkedIn, post reactions. Warmest contact at any company, dormant connections trending up, intro paths through the graph. The intelligence Affinity charges $2k/seat/yr for, on a Pi, across more channels than Affinity can reach.
Secure and compliant
Every inbound message screened before reaching the agent. Every agent action recorded in a durable audit trail. GDPR data subject rights — export and erasure — handled in conversation with confirmation gates and deletion receipts. Data stays on the device. The owner is the controller and the processor.
The category, by reference points.
The relationship-CRM category is being repriced upward by AI. Three vendors mark the corners of the space Maxy operates inside.
Folk
Seed (2021). Raised ≈ $3.3M. ~$8.3M ARR on 55 staff. Founder-led, content-driven, capital-efficient.
What this validates: founder-led sales pays.
Attio
Series B (Aug 2025). Raised $116M+ ($52M led by GV / Google Ventures). 5,000 paying customers. On track to 4× ARR in 2025.
What this validates: GV bets the next 25 years of CRM are AI-native.
Affinity
Series C (2021). Raised $120M at $600M valuation. 1,700+ customers in 70 countries. ~$2k/seat/yr. Quantified relationship-strength scoring for VC, PE, banking.
What this validates: warm-intro path-finding commands real money.
Affinity is the load-bearing comparable. The same primitive at consumer scale, on user-owned hardware, with federation, is a much larger market.
06Folk and Attio cannot reach where Maxy lives.
Affinity can — for one firm, on the cloud, at $2k/seat/yr.
| Folk | Attio | Affinity | Maxy | |
|---|---|---|---|---|
| Stage / raised | Seed / ~$3.3M | Series B / $116M+ | Series C / $120M | Pre-revenue |
| ICP | Founder-led sales, SMB | GTM teams, AI-native cos | VC, PE, banking | Founder-led sales (Pi-owners) |
| Pricing | $19–$159/mo | $69+/mo + credits | ~$2k/seat/yr | £280 hardware (one-time) |
| Storage | Cloud | Cloud | Cloud | Local Pi |
| Relationships | Inferred (heuristic) | Typed edges | Computed strength score | Typed edges + composite weights |
| Channels | Email, calendar, scraped LinkedIn | Email, calendar | Email, calendar, 40+ enrichment sources | Email, calendar, LinkedIn export, WhatsApp, Signal, Telegram, post reactions |
| Schema | Fixed | User-designed (high tax) | Fixed (CRM domain) | Opinionated default + extensible |
| Multi-hop traversal | No | API only, single multi-value hop | Yes (firm-bounded) | Native Cypher, unbounded |
| Temporal analysis | No | Limited | Yes | Native |
| LinkedIn ingestion | TOS-violating extension | None native | TOS-violating extension | Export + PDF (TOS-compliant) |
| Vendor's GDPR role | Processor | Processor | Processor | Tool vendor (no role) |
| AI privacy | Cloud LLM | Cloud LLM | Cloud LLM | Claude via user's OAuth; graph never leaves device |
| Cross-firm reach | No | No | No (firm-bounded) | Yes (peer-to-peer mesh)on the roadmap |
| Federation-ready | No | No | No | Yeson the roadmap |
Why none of them can pivot here.
Attio's $116M is committed to a cloud architecture. Their investors expect SaaS-scale margins, which require centralised data. Folk has no incentive to move local — their wedge is the content-led, lightweight CRM, not the privacy layer. Affinity's enterprise customers expect cloud control planes; the data-room of a VC firm is not relocating to a Pi. Each of them is structurally locked to their delivery model.
The local-first, federation-ready position is uncontested. Not by accident — the architectural decisions that produce it are the ones every cloud CRM cannot retrofit without rebuilding the company.
Five things Maxy does that
Folk and Attio structurally cannot.
Each is a consequence of architecture, not feature work. None of them can be added to a cloud CRM without rebuilding it.
1 · Quantified relationship strength
Seven-component composite score — recency, frequency, reciprocity, channel diversity, initiation balance, engagement depth, public endorsement — recomputed nightly across every edge. Folk has heuristics; Attio has nothing; Affinity has the primitive but only across one firm at $2k/seat/yr.
on the roadmap
2 · Multi-channel ingestion without exfiltration
WhatsApp, Signal, Telegram, email, calendar, LinkedIn export, Substack — ingested locally from on-device databases or user-controlled exports. The data never moves. No cloud CRM can touch the messaging channels where most relationship signal actually lives.
3 · Native graph traversal
Neo4j locally; arbitrary Cypher available. Queries Folk and Attio cannot express: “connections where strength is rising over the last 90 days but no message in 30”, “people who reacted to my last 5 posts but I've never messaged”, “dormant connections at companies in my watchlist”.
4 · Temporal awareness
Edge weights recompute nightly. The graph itself has a time dimension — whose relationship is trending up, whose is cooling, when a contact went dormant, what the warmest cohort of connections looked like a year ago.
on the roadmap
5 · Federation-ready architecture
Each Maxy node is a sealed local graph with a public-key identity. A peer-to-peer mesh allows nodes to broadcast intro requests to other operators without any graph data being shared. Architectural foundation for the network-effect moat — not retrofittable onto Folk or Attio.
on the roadmap
The advisor's
discovery-call follow-up workflow.
An independent advisor says: “After every discovery call, draft the recap. If the prospect signalled buying intent, send the proposal and update my pipeline. If they signalled later-stage, file under nurture and surface again in 30 days. Either way, log the relationship signal against the graph so the warmest contacts keep rising.” This workflow branches on a semantic condition, runs parallel paths, and updates a quantified relationship score. Real process, run daily.
Maxy reads the call record (transcript, calendar, prior thread) and drafts the recap. “I have drafted a recap for the Patels' discovery call yesterday — they asked about pricing twice and referenced a deadline. Shall I send the proposal?” The advisor reviews, adjusts tone, approves.
Replies route by signal. Buying intent triggers a proposal draft and a calendar hold for the follow-up call, both sent in the advisor's voice. Later-stage triggers a nurture queue entry with a 30-day resurface. The advisor gets a morning summary: “3 recaps sent. 2 buying signals — proposals drafted. 1 nurture — resurfacing on 26 May.”
Morning briefing includes the overnight digest, a list of warm connections trending up across the week, and any commitments coming due. The advisor's time shifts from admin to high-value work — thinking, advising, networking.
Six roles. One operator.
Zero dilution.
Admin, scheduling, research, content, coaching, and the invisible work of keeping the knowledge graph current. Maxy handles all six so the thinking stays uninterrupted.
Project Manager
Projects, tasks, dependencies, sessions, lifecycle. The operator-facing memory of what is happening now.
Personal Assistant
Scheduling, admin, messaging, browser control, morning briefings. The interface to every surface the operator already uses.
Research Assistant
Web search, knowledge retrieval, images, citations. Answers that cite their sources.
Content Producer
Documents, PDFs, ingestion, visuals. Turns operator thought into artefacts the operator would have written.
Coach
Accountability, patterns, progress, reviews. The role that keeps the operator honest about their own follow-through.
Database Operator — the foundation
Ingestion, entity resolution, edges, schema, cleanup. Turns every captured surface into a node in the graph and keeps every edge current. The role that makes every other role possible.
The moat is architectural.
Four axes. No shortcuts.
GBrain
- Packaged for non-technical operators. GBrain runs behind OpenClaw, Hermes, and operator-maintained cron. Maxy is Claude Code wrapped for non-technical operators, reached through chat.
- On-device, not cloud Postgres. GBrain runs Postgres on a developer machine or cloud. Maxy ships the graph, vector index, and agent on a private appliance the operator owns.
- Graph-first, not Markdown-first. Markdown-with-Git is a developer's personal rig. Maxy persists to Neo4j with vector embeddings — read through conversation, not a file browser.
Anthropic
- Anthropic builds Claude Code for developers. Its power comes from explicit primitives — skills, tools, hooks, subagents, MCP servers. Maxy wraps all of it into conversation for non-technical users.
- Claude Desktop is where Anthropic is most active in the agentic space — dispatch, computer control, consumer-facing AI. We monitor this closely.
- The moat is specific: natural language as runtime, bespoke workflows underpinned by an IC's private connection graph. Everything else, Anthropic can do. These three together, they cannot — not without building an operations layer for the individual contributor, which is not their product.
Founder uses it daily. Channel partners committed.
Three independent proofs. The founder is the clearest IC inside the product, day after day. A channel partner has put real money down against a vertical built on the same platform. Direct-to-IC subscription is the next commercial path to open.
Founder as IC — daily user
Joel built Maxy as a heavy daily Claude Code user, then wrapped it for non-technical operators. He cancelled his claude.ai subscription. This investor deck was created end-to-end through Maxy. The product is sharpened by the IC who lives inside it.
Real Agency — channel partner
Joint venture with eXp network agents. Vertical built on the Maxy stack under a separate brand. Core workflow and price points validated with Alex (solo agent) and Adam Mackay (multi-agent channel partner). Steve Backley OBE and Roger Black directly engaged as platform contributors. realagency.network →
Real Lettings — sponsor and pilot cohort
Dan McLeod (30 years UK lettings, recently exited Foxtons) co-designed the wedge over a 60-minute process-mapping session. Five-agent pilot cohort identified, all running Reapit, 1,000+ tenancies combined, ~50+ tenant calls/day expected aggregate. The thesis: preserve the human-to-human first call as relationship capital, automate everything downstream.
Direct-to-IC subscription is the primary commercial path. Channel partners (Real Agency first, Real Lettings second) are the complementary lane, funded by partner capital, riding on the same Maxy stack.
13High-margin software.
Primarily license revenue.
ICs pay an annual license. Users pay their own AI costs directly to providers. Channel partners ride on the same stack under their own brand and fund their own go-to-market. Software scales horizontally with minimal incremental cost.
Direct IC Subscription
Annual licenses for the individual contributor. Maxy Pro at £1,999/year with hardware extra, or £2,500 lifetime for the first 100 (Founding 100). Verticals (Real Agent, Real Lettings, Professional services) priced separately and higher than base Maxy Pro, reflecting the calibrated workflow library per vertical. Zero ongoing infrastructure cost — users pay their own Claude Max 20X subscription directly to Anthropic.
Channel Partners
Vertical brands built on the Maxy stack — partner-funded go-to-market in exchange for revenue share with exclusivity. Real Agency is the first; the model is repeatable into any vertical with relationship-driven sales. Complementary to direct subscription, not a replacement.
Founding 100
First one hundred IC operators get lifetime Maxy Pro at £2,500 in exchange for shaping the product. Direct line to the founder, weighted input on roadmap, founding cohort convening. The cohort closes at 100. Real Agent runs its own separate founders allocation (100 offices, £5,000 lifetime Solo, £15,000 lifetime Office) for the estate-agent vertical.
Unit Economics
Low marginal cost: hardware provisioned and dispatched once, then ongoing AI costs borne by the user. Each channel partner funds their own go-to-market. Software scales horizontally with no incremental cost.
Joel Smalley.
Polymath · Builder · Bootstrapped
Career
- 24 years in capital markets — JPMorgan, CIBC, Daiwa. Built 3 proprietary trading desks at tier-1 institutions.
- MBA, Dean's List — Rotman School of Management. Studied integrative thinking under Roger Martin.
- Supermoney — enterprise DLT for BMW Finance, Volvo Finance, Poste Italiane.
- Independent research — 500K+ reads on ResearchGate. 32K Substack subscribers.
Why this, why me
- Built Maxy as a heavy daily Claude Code user — realised no non-technical user would ever access that power through Anthropic's own products. Built the bridge myself.
- I use it. Cancelled my claude.ai subscription. This pitch deck was created entirely via Maxy.
- Ship velocity: new product every month since 2025. Zero-to-prototype in 10 days.
- Cambridge EduX Hackathon 2025 winner (beat 27 teams).
GBP 250,000 to scale what is already working.
Private Network Intelligence, shipped.
We are not building another AI dashboard. We are building the operations layer for the individual contributor — private memory, quantified relationship strength, multi-channel ingestion, and bespoke workflows, running on hardware the operator owns. The IC always stays competent. The IC always stays in control.
The category is real. Affinity proved that warm-intro path-finding commands real money inside one firm. Maxy ships the same primitive at IC scale, on user-owned hardware, across more channels, with federation in the roadmap. Channel partners (Real Agency the first) ride on the same stack under their own brand. This is product-market fit built by an IC who lives inside the product every day.
joel.smalley@rubytech.llc
getmaxy.com