Substrates that know your business

We make AI an asset. Literally.

What if one person could run a business? Look at our Showcases →

Atteste is a SaaS company. It runs day-to-day with one human and one AI agent. Practacular is a tax-practice multiplier — live, paying customers. Same pattern, different vertical. What both share is a custom-fit knowledge layer that gives AI access to everything the business knows. We call it a Substrate — and at handover it's a line on your balance sheet, not a bill.

AI is everywhere. Almost none of it knows your business.

Generic LLMs are powerful tools — they can write, code, summarise, reason about general knowledge. But the questions a business actually pays AI to answer aren't generic.

They're about your project files, your contractor history, your tax records, your customer conversations, your accumulated decisions. None of that is in the model's training data. And none of it gets useful unless someone designs the path between your business's specific knowledge and the AI's general capability.

That path is the Substrate. The gap between generic AI and useful AI is what we close.

What we mean by Substrate

A custom-built knowledge layer. It captures what your business already knows — documents, workflows, decisions, conversations, system records — into a queryable structure that lives on hardware you control. AI connects to it via MCP, the open Model Context Protocol. Once connected, the same generic LLM that struggles with your business becomes fluent in it.

The CDC lockup exploded into its three layers: navy core, copper rails, nameplate

Layer 01 — the navy core 01 — Capture

Ingest what your business knows

Files, decisions, workflows, customer conversations, system records — pulled into a queryable structure on hardware you control. Self-hosted by design. Owned by you from minute one.

Layer 02 — the copper rails 02 — Connect

Speak MCP, bring your own LLM

The Substrate exposes itself via the open Model Context Protocol. Use Claude. Use GPT. Use Gemini. Use whatever ships next. Model-agnostic so you don't lock to a vendor.

Layer 03 — the nameplate 03 — Multiply

Operate at AI speed, sign at human speed

Operators query in plain English. Drafts go to an outbox. Compliance gates stay where they matter. Day-to-day operations run at AI speed; the human signs the work that needs a signature.

Reassembled

Together, it's simply your business — running.

Three layers, machined to fit one company: yours. The diagnostic maps which layers you already have and which we build.

Begin the diagnostic

§ 03 — In production

We build Substrates. Three examples.

Each Substrate fits a specific business and a specific operating model. The three we've shipped — all in production — are deliberately different. Atteste operates a software company end-to-end with a single human in the loop. Practacular multiplies the productivity of an accounting practice. BlitzBox is the productized version — a single-tenant local-first substrate on a Mac Mini at your premises. Same underlying pattern. Different verticals.

Atteste — the substrate handles

Marketing campaigns
Sales outreach & sequences
Financial reconciliation
Customer support
Founder calendar & outreach

Practacular — the substrate handles

Tax review workflows
Compliance lookups
Client engagement records
Workpaper preparation
Practice-wide RAG

BlitzBox — the substrate ships as

Mac Mini on your premises
Brain / Financial / Email / Compliance
Mothership-rotated credentials
Local-first, POPIA-shaped
R30,000 + R15,000/mo

Showcase A — Operator-substrate

One operator, one AI, one company.

Atteste is an art collection management platform. The product runs publicly at atteste.art. Less visible: the company behind it operates as a single human and an AI agent. Marketing campaigns, sales sequences, financial reconciliation, customer support — handled day-to-day by Claude on a Substrate that captured the entire business.

The operator's role is compliance, judgement calls, and the things that genuinely require a human signature. The work that doesn't is delegated to the substrate. This is what becomes possible when AI gets proper access to a business's knowledge — not an experiment, a working operating model.

Visit atteste.art → Read the write-up →

command-centre / video-studio

Showcase B — Practice multiplier

Advanced RAG over an entire accounting practice.

Practacular is AI practice management for South African accounting firms. Live, paying customers. Its substrate is an Obsidian graph deployed across the practice's tax reviews, compliance workflows, and client engagements — advanced semantic RAG plus flat grep-RAG, queryable in plain English by every accountant on the team.

Tax review cycles compress. Compliance lookups stop being manual. The practice operates at higher leverage because its accumulated knowledge becomes instantly accessible — to every accountant, on every engagement, in every conversation with a client.

Visit practacular.com → Read the write-up →

practacular / firm-brain · graph

Showcase C — Productized substrate

The substrate, in a Mac Mini you can hold.

BlitzBox is the productized version of what we do for bespoke clients. A single-tenant local-first AI substrate that ships on a custom-finished Apple Mac Mini and runs at your premises. Brain, Financial, Email, and Compliance modules always on. Your business data stays on your hardware.

We sell it directly at R30,000 upfront + R15,000/month, with a refundable R5,000 EFT deposit to reserve a commissioning slot. Karel personally provisions each Mini at our Cape Town master station before it ships.

See BlitzBox → View pricing →

You end the engagement owning a capital asset.

At handover you receive Operational IP — codified ontology, runbooks, decision history, integration map, audit log, retrieval graph. MIT-licensed. Self-hosted. Capitalisable on your balance sheet. Transferable in M&A.

This is not a consulting receipt. It is a thing you buy — fixed scope, fixed acceptance criteria, owned by you on day one. The Handover Manifest is the literal deliverables list. See the Handover Manifest, the three valuation mechanisms, and the math →

How we work

Four principles. They're load-bearing — every engagement we take on is gated by them.

Tenet 01

The substrate is the product

Generic AI is everywhere; what's rare is AI that knows your business. We don't sell the AI. We design and build the substrate that makes the AI useful to you specifically. Your edge is custom; our discipline is building it.

Tenet 02

Model-agnostic by design

The Substrate exposes itself via MCP — the open Model Context Protocol. Use Claude. Use GPT. Use Gemini. Use whatever ships next. The substrate doesn't care, and we won't lock you to a vendor we don't run.

Tenet 03

Owned by you

Substrate code is MIT-licensed to you from minute one. Self-hosted on hardware you control. No data egress to our infrastructure. If you stop working with us, you keep the substrate, the audit log, and the runbook. We don't hold your context hostage.

Tenet 04

Humans where they matter

Audit log captures every command. Outbox gates every draft. Credentials stay with you — the substrate refuses to type passwords or 2FA. Day-to-day operations run at AI speed; the human signs the work that needs a signature. We never auto-fire.

Who

Context Design Co operates under Yield SPM (Pty) Ltd — a South African studio building AI-multiplied businesses. The two we ship in production — Atteste and Practacular — are our own.

We use the same patterns we sell. The substrate-design discipline you see on this page didn't come from a whiteboard; it came from running our own businesses on it.

We're the studio that does it because we already do it.

Operating across Cape Town · Stellenbosch · Bloemfontein · Johannesburg