Kanon installs the structured data substrate that AI was missing. A markdown-native repository your team owns, that agents can finally read, write and reason over without inventing.
~/data-substrate ├── canon/ # voice, audience, principles ├── memory/ # what the business knows ├── decisions/ # append-only record ├── operations/ # meetings, contacts, tasks ├── customers/ # per-account working set └── CLAUDE.md # the kernel $ kanon ask "what did we promise the Mueller account in Q1?" → 3 commitments found in operations/meetings/ → all citations linked. no fabrication.
Sectors we serve
The problem
Software companies already run on structured data. Everywhere else, knowledge lives in heads, habits and filing cabinets. AI on top of that answers fluently and wrongly. The model isn't broken. The substrate underneath is.
Point an LLM at a Drive sprawl and you'll get fluent, plausible, often wrong answers — with no audit trail your operations team can defend.
Retrieval assumes structured source material. Most non-software businesses don't have it. Adding RAG on top of unstructured chaos amplifies the chaos.
You leave the engagement with a strategy PDF and the same disorganized Drive. No durable infrastructure. No system the next agent can read.
The product
We design and install a structured data substrate: a markdown-native, git-tracked repository your team owns. Every claim traces to a file. Every agent reads the same canon. Nothing is invented.
Structured interviews surface how your business writes, decides and draws lines. The output becomes canon/ — the layer every agent reads first.
A kernel (CLAUDE.md), routing rules, memory, decisions, and one-home-per-insight discipline. Structure tells agents exactly where to look.
Meeting processors, research, customer onboarding, competitor tracking, weekly digests. Each grounded in the substrate. Each accountable to a file.
Markdown source, git-tracked. No vendor lock-in. Export, fork or self-host whenever you want. The data layer is yours on disk, in plain text.
Visible "I don't know" states. Citation-to-source links on every answer. The system refuses to invent — epistemic honesty is built in, not added later.
EU-first by default. Local-AI and on-prem variants available. Build with compliance baked in, not bolted on after the audit.
How it works
A focused session with founders, ops leads and the people closest to the data. We surface voice, decision style and the questions your team actually asks AI.
We scaffold the repository: canon, memory, decisions, operations, customers. Routing rules and the kernel that orchestrates every agent.
Real meetings, decisions and customer notes flow into the substrate. Your team writes; agents read. The data layer densifies in hours, not weeks.
First production agents go live on top of the substrate. Handover, training, and a maintenance cadence so the data layer evolves with your business.
Pricing
One substrate. As many agents on top as your business needs. Transparent recurring fees, no per-seat games.
The data substrate, installed.
When the substrate starts paying for itself.
Multi-team, multi-region, regulated industries.
Prices in EUR, excl. VAT. Talk to us about volume, pilots, and academic pricing.
Our approach
Plumbing is commoditising fast — every vendor will have RAG, every vendor will have agents. The durable advantage is the structure they run on. We build that structure first.
Markdown on disk, git history intact. No proprietary format, no vendor lock-in. If we disappeared tomorrow, your substrate would keep working.
Every claim traces to a file. "I don't know" is a first-class answer. The system refuses to invent — by design, not by prompt-engineering.
Before structure, voice. We capture how your business writes and decides — so agents respond as you would, not as a generic AI on top of your data.
FAQ
Consultancies ship strategy slides. We ship infrastructure. At the end of an engagement you own a working repository, not a PDF. Implementation is the beginning, not the end — we maintain the substrate against drift, schema changes, and new tooling.
RAG assumes the source material is already structured. For most non-software businesses, it isn't. We create the structured substrate first; retrieval and agents on top then become a solved problem rather than an unsolved one.
No. The repository is yours, on git, under your account. We can host or operate it for you on request, but the default is full customer ownership and zero vendor lock-in. Export is trivial — it's already plain text.
The architecture is DSGVO-aware from day one: data minimization, clear retention, append-only decisions, audit trails. For regulated industries we offer local-AI and on-prem variants. We're happy to walk through your specific posture before signing anything.
Two days from kickoff to a working substrate with first production agents. The discovery session on day one is where the durable value lives — by day two you're operating on top of it.
Non-software SMEs with five to two hundred people, typically. Mittelstand sweet spot. If you're a SaaS company already running on structured data, you probably don't need us.
Tell us about your team, your data, and what you've already tried. We reply within one business day with a fit assessment and next steps.