be in the knowing
havn is a collection of tools, workflows and orchestration focusing on automation and AI.
Four ways in.
One place to stand.
Each tool has a clear job. Together they form a workspace where material can move from knowing to making without being copied between disconnected apps.
Knowing
The layer that makes a canvas queryable in plain language.
Multimodal extraction, local BM25 + vector retrieval, embeddings, and board bridges that turn what you put down into usable memory.
The living features
Made to move,
not just show.
These aren't static screenshots. Each product carries a small live behavior that hints at how the whole system feels to work with — context in motion, people present, compute close by.
Knowing — a canvas that knows
Put anything on it, then ask it anything about what you put there.
Board — made to be shared
A surface people step into, with the work staying visible while the system works beneath.
Kernel — compute close by
The expensive work moves to the right local runtime, so the making layer stays calm.
The surface stays calm because the foundation is deep.
Havn is built as a system of boundaries. The UI does not need to know how every primitive works. It can stay focused on helping someone think, while the layers beneath it take care of the hard parts.
Board
the knowing canvas
Knowing
multimodal memory
Viewer
spatial runtime
Kernel
compute foundation
Built in the open,
with receipts.
The product is the experience. Each supporting layer has its own page — and the repo behind it, for when you want to read the evidence.
The people behind
the system.
Havn is being shaped by a small team that cares about the details: useful tools, durable foundations, and a calmer way to work with complexity.
Generative AI makes the thing for you. havn makes you better at making the thing — because it understands your material.
The intelligence is there to surface context, remove the tedious steps, and keep the author in control.
Bring the material
you actually work with.
We are building havn for people with real, messy, revisited material — not another feed, dashboard, or prompt box.