Today, 3 October 2026, we are opening a preview of the Alkemata Knowledge Map. It is our first step toward making “Understand” a place you can explore: a landscape of topics, connections and questions, where useful discoveries can remain available for the next person.
The map grows out of Alkemata’s three guiding verbs: Explore, Understand, Shape. Discover an idea that matters. Work through it until you can explain and question it. Use what you learn to do something. AI can help throughout this process, but human curiosity, judgement and responsibility give it direction.
Explore: find a question worth following
Exploration begins before we know exactly what we are looking for. An unfamiliar invention, a surprising scientific result or a problem in everyday life can interrupt our habits and make us curious. Alkemata’s articles and press reviews are well suited to this moment: they introduce ideas, explain why they matter and connect technical developments to human consequences.
The aim is not to maximise the number of things you read. It is to help you find something worth your attention. What is promising? What could go wrong? How does this affect people? Which question would be worth investigating further?
Your own AI can help compare articles, identify assumptions or connect a new idea to your interests. But discovery also needs room for surprise. A system that only returns what it thinks you already like can narrow exploration. Human editorial choices and unexpected connections remain valuable.
Understand: build a route through knowledge
An interesting article is often a beginning. Understanding may require several conversations, a calculation, a diagram, a source check and a return to something more basic. We move forward, take a detour, discover a missing prerequisite and try again.
A conversation with a language model is useful for this work. You can ask for another explanation, challenge an answer or change the level of detail. Yet a long chat records the order in which you happened to ask questions. Someone else may have a different starting point, and the useful results can become buried among repetitions and wrong turns.
We want to preserve the discoveries and the routes that connect them. A good exploration should leave traces: a clear explanation, an unanswered question, a checked calculation, a source, or a connection that another human or AI agent can follow.
Why a map? Knowledge has branches and meeting points
Knowledge rarely fits a single sequence. One topic can be reached from several directions, and an idea can belong to more than one field. Entropy, for example, connects physics and information theory. AI memory connects software design, learning, privacy and human agency.
In the map’s intended structure, topics are locations, connections link them, and learning trails describe routes through them. A trail can branch. Two trails can meet at the same topic. A reader can follow an established route or leave it to investigate a neighbouring question.
The landscape provides a visual frame: regions for broad areas, towns or buildings for more specific subjects, and paths or bridges for connections. Zooming lets the visitor move between orientation and detail. Selecting a topic opens its explanation without making every part of the landscape compete for attention at once.
The geography is a metaphor. Distance on the map does not prove that two ideas are similar, and a bridge does not establish a scientific relationship. A useful connection needs a stated meaning: “requires this concept,” “applies this method,” “offers an alternative explanation,” or “raises this question.”
A landscape to remember
There is also a memory-palace possibility: placing ideas in recognisable locations and revisiting them through a familiar route. A courtyard could hold a foundational concept; a workshop could contain an experiment; a bridge could remind you of a connection between fields.
This is a design direction, not a claim that the preview has demonstrated better learning or memorisation. Stable landmarks may make the experience easier to revisit. The map should grow without continually moving the places people have learned to recognise. A personal memory-palace arrangement is a possible future view of the same shared knowledge.
AI as a guide—and as a contributor
Alkemata does not need to replace the capable models readers already use. The aim is to give a reader’s chosen AI useful material and a clear structure to work with. You should be able to carry a question, sources and a checkpoint into another conversation, then bring useful results back.
A model could help you choose prerequisites, explain a difficult step, compare competing arguments or suggest a route toward a goal. It could also test understanding by asking you to predict an outcome or explain an idea in your own words. A convincing answer from the model is not, by itself, evidence that either the answer is correct or the learner understands it.
For contributors, the ambition is equally concrete. Humans and authorised agents could propose topics, attach sources, improve an explanation, identify a contradiction or connect two existing locations. The contribution should remain inspectable: who produced it, what evidence supports it, what changed and what still needs review.
The visual map and the machine interface should describe the same underlying topics. An agent should be able to find a topic by its identity and relationships; it should not need to interpret a screenshot or guess which building contains it. This also allows the visual presentation to evolve without breaking existing references.
Shape: turn understanding into something testable
Understanding becomes especially valuable when it changes what we can do. “Shape” is the point where a question leads to an experiment, a prototype, a public proposal, a practical decision or an improvement to an existing system.
The scale can be small. A reader exploring AI memory might first understand retention and provenance, then draft a memory policy and test whether it behaves as intended. Another reader might follow a technical trail and build a simulation that makes an assumption visible.
The results should reconnect to the knowledge that made them possible. An experiment may support an explanation, reveal a limitation or generate a new question. Explore, Understand and Shape therefore form a continuing cycle: action gives us new material to investigate.
What opens today—and what remains an ambition
The preview at map.alkemata.com offers an early knowledge landscape with a zoomable map, topic search and published topic material. Its public topic index includes subjects such as AI Evaluation, Agentic Systems and Algorithmic Accountability. This is a starting point for exploring the experience, not a complete atlas of knowledge.
The application’s capability description advertises MCP access, including write support, and states that drafts require editor review. That describes an interface and a publishing boundary; it does not mean an anonymous visitor or every AI client can immediately contribute. Compatibility, identity and permissions matter.
Today’s preview does not establish that every part of the longer-term vision is complete. Rich branching trails, portable conversation checkpoints, personal memory palaces and a mature community of human and agent contributors are directions to develop and evaluate. We will distinguish what visitors can actually use from what we hope to build.
The map also has its own interface. Do not assume that the existing Alkemata article MCP connection can edit it: the article service is described separately in Alkemata for AI: What Capsules, llms.txt and MCP Are Actually For.
Try the preview with one question
- Open the Alkemata Knowledge Map and choose a topic that makes you curious.
- Move between the landscape and topic search. Notice whether you can find your way and return to a location.
- Read an explanation, then ask your own LLM to clarify a difficulty or identify a prerequisite. Share the relevant text or link; check that the model can actually retrieve it.
- Keep a short record of what became clearer, which sources you checked and what remains unresolved.
- Tell us which connection, explanation or route would have helped you continue.
The most useful feedback is specific: “I reached this topic, but needed that concept first”; “these two subjects should be connected”; or “this explanation leaves an important assumption unstated.” Such observations help turn a visual landscape into a place where understanding can accumulate.
Explore invites us in. Understand helps us find our way. Shape gives the journey consequences. With today’s preview, we begin testing how a shared map can support that journey—and how humans and AI can leave useful paths for others to follow.
