How it works
Curated context
Documents record what was decided. They don't record why, or what was considered and rejected, or what changed two weeks later. By the time knowledge makes it into documentation, the thinking that produced it has already moved on.
Curated context starts earlier — at the moment of thinking, not the moment of writing.
Workflow · Create a goal
Create goal
Write a goal directly to your workspace
Ship unified design-system tokens by end of Q2
Replace ad-hoc Tailwind utilities with named primitives + tokens across platform-ui, so the marketing site can adopt the same look without copy/paste.
tokens.css covers shadows + spacing + type
tokens defined → 20 8
kitchen sink uses .eg-* classes exclusively
inline-style usages → 0 12
Authored, not extracted.
Engramic's knowledge graph is built from typed topics: goals, constraints, decisions, concepts. Each one is authored deliberately — by a human, by an agent, or by both working together. Not pulled from a document corpus. Not inferred from usage patterns. Put there intentionally, with the reasoning attached.
That distinction matters. An extracted knowledge base tells you what your organisation has written. A curated knowledge graph tells you what your organisation has decided, what it's committed to, and what it's still working through. Those are different things, and agents need the latter.
Knowledge that knows what time it is.
A knowledge base is static. Engramic's graph has a sense of time. Temporal events record what happened, when, and why — layered on top of the topics they relate to. When a goal changes, the change is recorded. When a decision gets revisited, the prior version is preserved alongside the new one.
This means the graph accumulates not just what the organisation knows, but the history of how it came to know it. When something conflicts with what was previously agreed, it surfaces — rather than quietly undermining an agent that had no way of knowing things had changed.
Conflicts appearing in the graph aren't a sign the system is broken. They're a sign it's working.
- 16:48Q3 forecast committed — 18% over planPlan Agent
- 14:32Conflict surfaced — Tier B vs Q3 planAgent
- 14:22Pricing tier A revisedSarah Mitchell
- 11:05How will Tier B affect annual contracts?James Kim
- 09:42Initial pricing model committedFinance Team
- 09:15Tier B raised — 12% uplift agreedMaya Patel
AI Fluency · 4D Framework
A practical model for working with AI deliberately — referenced in our own approach to authoring context.
Delegation
what to ask of AI
Description
how to frame the task
Discernment
what to trust in the answer
Diligence
staying accountable for the outcome
The platform works best when the people using it do too.
Curated context is only as good as the conversations that produce it. That means the humans authoring knowledge need to be comfortable working with AI — not just using it, but collaborating with it deliberately.
The 4D AI Fluency Framework — Delegation, Description, Discernment, and Diligence — developed by Prof. Rick Dakan and Prof. Joseph Feller in partnership with Anthropic, is a practical starting point. It's free, model-agnostic, and designed for exactly this: helping people develop the judgement to work with AI effectively, not just efficiently. AI Fluency: Framework & Foundations is available on Anthropic Academy →
Engramic doesn't replace that fluency. It gives it somewhere to land.
Knowledge improves by asking questions.
Engramic's own agents are designed to help fill knowledge gaps — not just for agents being onboarded, but for the humans building the graph. When a topic is underspecified, when a goal lacks a constraint, when two decisions look like they might conflict — Engramic asks. Socratically, specifically, one question at a time.
This means the knowledge graph doesn't just grow as work gets done. It gets sharper. The act of answering a well-formed question surfaces knowledge that would otherwise have stayed in someone's head — and puts it somewhere agents can use it.
12 open · 8 answered (24h) · 5 topics sharpened
How does this interact with existing annual contracts?
Triggered by Tier B raised — 12% uplift. The graph has 23 active annual contracts on Tier B. Mid-term uplift may conflict with the renewal clause in the standard MSA.
Once the knowledge is in the graph, the next question is how it reaches an agent before it acts.
That's what agent onboarding does