Every AI tool you use is only as good as the context it has about your business. KnowledgeForge is the context-engineering retainer that turns scattered company knowledge — documents, decisions, terminology, customer history — into a working layer your AI agents and assistants can actually use. It is not knowledge management software. It is the practice of keeping the context your AI sees correct, current, and queryable.
Per knowledge worker at £150/hr, lost to re-explaining business context to AI tools. A team of five is £1,875/week before any KnowledgeForge work.
Reported lift in AI coding-task accuracy from a properly written context file. Same pattern holds for any AI agent given a real working context.
Context layers compound. At 1 month they're useful; at 3 months they're load-bearing; at 12 months they're a moat your competitors can't copy.
Most businesses already tried knowledge management. They bought a wiki, asked staff to "document things," and watched the wiki go stale inside a quarter. The mistake wasn't laziness — it was the model. A wiki is a place for humans to read; what your AI agents need is a working memory they read at the start of every run.
The phrase comes from production AI practice — practitioners shipping agent systems in 2026 named the missing layer "context engineering" once it became clear that prompt tweaks couldn't substitute for it. The model is commodity; the context around it is not.
We never start with Build. Every engagement begins with a fixed-price Discovery so you see what you're buying before the meter starts. The Build stage takes two weeks inside the business. The Maintain stage is the monthly retainer that keeps the context layer correct as your business changes — which it will.
Stage 01
Two weeks inside the business, mostly remote. We sit with the people who hold the tacit knowledge, map what exists, identify the gaps, and write the build plan.
Stage 02
Two weeks inside the business. We build the actual context layer to the plan from Discovery, train the team on the governance pattern, and ship the eval framework that catches drift.
Stage 03
Monthly retainer. We keep the context layer current as your business changes — onboardings, departures, new products, new clients, new processes — and run the evals that catch drift before your AI starts giving wrong answers.
All prices exclude VAT. Discovery is mandatory before Build. Maintain is optional, but every business we've shipped Build to has moved onto it — context layers go stale without a retainer faster than any other artefact we deliver.
"Context" sounds vague until you see what's inside it. KnowledgeForge ships six concrete layers — each one a file or graph the AI reads, each one with a named owner inside the business.
Typed nodes for people, projects, customers, organisations, decisions, deadlines. Backlinks. The "who, what, when" your AI needs before it can answer anything specific.
The decision log your AI reads at session start. Corrections, preferences, locked architectural choices, things you wish you hadn't done. The brain's index.
Your business has its own language. The acronyms, the nicknames, the project codenames, the "we always call this X" rules. Captured once, applied everywhere.
50–100 hand-labelled real examples. Run weekly. Catches drift before your AI starts being wrong in ways no-one notices for two weeks.
RAG, GraphRAG, or maintained wiki — the right answer depends on your corpus. We pick the strategy and implement it; we don't sell you on whichever vendor pays best.
DPA, DPIA template, data-classification map, retention policy. The paperwork that turns "we use AI" into a defensible answer to your auditor.
Before we sell this to anyone else, we run it on ourselves. Every client engagement, every proposal, every internal decision Launchpad makes is captured into the same kind of context layer KnowledgeForge ships. That's how we know what stays current, what goes stale fastest, and what the retainer actually has to cover.
A live Obsidian vault, an auto-memory layer for our Cowork sessions, a per-project file structure that mirrors the entity graph, and an eval pass on every session-end sync.
If we tell you "this stays current with a £2,500/month retainer," it's because we've measured exactly how much it costs us to keep ours current. Not estimated. Measured.
It contains everything your AI sees about your business — clients, decisions, terminology, history. We treat it with the same care any internal system carrying that data would get. 72+ years of combined experience across NHS, Police and MOD, including SC-cleared personnel for regulated-sector work.
If yours isn't here, the contact form gets to a human inside one working day.
Notion and Confluence are tools. KnowledgeForge is the practice that turns the tool's contents into a layer AI can read. We're tool-agnostic on the storage — if you already have Notion, we work in Notion. If you've got SharePoint, we work in SharePoint. The deliverable is the structure, the governance and the eval framework, not the software licence.
If you don't already have a place for it, we'll usually recommend Obsidian — local-first, file-based markdown, no per-seat licence, and exactly the structure a context layer needs. But that's a recommendation, not a lock-in.
Three things in fixed proportion. A third pays for the weekly automated eval run and drift report. A third pays for monthly entity-graph refreshes against new sources (new clients, new staff, new projects, new decisions). A third pays for human time — quarterly governance review, two hours of office hours a month, ad-hoc questions when something changes faster than a quarter.
What it does not pay for is "we build new agents for you" — that's a separate engagement, either Launchpad Agent or a custom build.
KnowledgeForge is the layer underneath. Launchpad Agent is the layer on top. The agent reads from the context layer; the context layer is what makes the agent stop sounding generic and start sounding like your business. We often ship them together — KnowledgeForge Discovery + Build first (4 weeks, £6K), then Launchpad Agent Team (£4,995 + £599/month), then both retainers run in parallel.
Buying the agent without the context layer works for narrow workflows. For anything that touches client knowledge, internal terminology, or past decisions, the agent is a fraction as effective without the context layer behind it.
At month one you have a working layer and the eval framework catches the first round of "AI getting things wrong." At month three the layer is load-bearing — staff stop re-explaining basics to AI tools, and onboarding new people gets materially shorter. At month twelve it's a moat — your AI knows things about your business that a competitor couldn't reproduce without the same year of work.
The honest answer on payback: most of the cost is recovered inside six months at a 10-person business. Bigger teams compound faster. Smaller teams should consider whether they need Discovery + Build alone without the retainer.
That's exactly when the retainer matters most. Static businesses can sometimes get away with Discovery + Build and quarterly self-maintenance. Businesses with frequent change — new clients monthly, new staff quarterly, evolving services — see the context layer go stale fastest and benefit most from the retainer.
The eval framework is the early-warning system. If accuracy drops on the weekly run, we catch it before your AI starts making confident wrong answers in client-facing situations.
Yes. We map every source of context against your classification scheme as part of Discovery — what's public, what's internal, what's confidential, what's restricted. The build respects those classifications: restricted data stays in restricted stores, the agent only reads what its scope permits. The data-classification map is one of the documents you take away at the end of Build.
For regulated-sector work (NHS, financial services, central government, MOD-adjacent), the engagement runs under a per-project DPA and the lead is SC-cleared.
90 days' notice, no exit fee. You own the entity graph, the MEMORY.md, the terminology layer, the eval dataset and the DPIA. We hand them over in portable formats — Markdown for the prose, JSON for the eval data, plain SQL or graph-export for the entity store. No proprietary container, no vendor lock-in. If you need a 30-day hand-over period to a successor team, we'll do that too.
Book a 45-minute discovery call. We'll walk through what your AI currently doesn't know about your business, and tell you whether KnowledgeForge is the right answer — even if the honest answer is "not yet."
Book a discovery call