the agent gets the data. the human gets the picture.
A datalink is a picture, a web link, that carries the data. Your reader sees the picture; your AI reads the numbers out of the link.
The same address, read as numbers: /draw.json · open it in the explorer
Your AI writes the piece with the data in hand. What it leaves behind is a picture — and the data went into the image and did not come back out of it. Next month, whoever reads the piece has the picture and nothing else.
the picture
A chart can show you the shape of a hundred numbers. It cannot hand you the hundred numbers.
the model
Given the picture, a model reconstructs the values by looking. It is good at that — and it is still a reading, and it cannot tell which parts it read and which it filled in.
the artefact
The piece ships; the context that made it does not. Whoever reads it next — a person, an agent, you next year — has the picture, and no data to compare it to.
One address, carrying the data itself. Open it and you get the picture; read it and you get the numbers — they are right there in the address, exactly as measured. There is no second version of the truth to keep in sync, because there is only the link.
one
An ordinary web address. It carries the data — not a pointer to a page, not a screenshot, the data.
two
What a reader sees. A chart, a diagram, a map — drawn from that data, every time.
three
What an agent reads. Already in the address — exact, quotable, no looking involved.
a reader · the picture
loading…
Everything the reader needs, at a glance.
your AI · the link
loading…
Your agent changes one parameter and gets a different picture — no redrawing, no exporting, no asking anyone. Try it: change a number above and press Enter.
The link is the artefact, and it serves both — no translation step, so nothing is lost.
the human · gets the picture
the agent · gets the data
The agent writes the piece — a README, an issue, a pull request, a post — and where it needs a chart it composes a link instead of pasting an image. The report looks the same as ever; what changed is that every chart still holds its numbers.
Write our Q3 review — signups, churn and revenue by month, with a chart
for each. Use datalinks for the charts, and keep the links, so we can pull the
numbers out of them later.
what the reader gets
Charts from one renderer, the same every time — not whatever the agent improvised that day.
what stays behind
Every picture is a link with the numbers inside it. Nothing was flattened on the way.
what it means later
Six months on, an agent with memory reads the link and finds the intent and the numbers — not a picture of them.
A picture is a file. Files have to be re-made, re-uploaded and committed again — and a repository full of them gets heavier every time a number changes. A link is text. Text just changes.
case one
Today. The number changes, so the chart is re-rendered, re-uploaded and committed as a binary. The repository grows, and the diff says “image changed” — not what changed.
With a datalink. The agent changes one number in the URL and commits. The diff shows the number, the picture follows it, and a year later the history still says what you meant.
case two
Today. Six languages, six chart files, six renders, six commits — and every change to the data means doing it six times.
With a datalink. The language is a parameter. Six charts, six URLs, one line of text each — generated from the same source by the same agent.
The same diagram in six languages, six links — the language lives in the labels. No six files, no six renders, nothing to keep in sync.
The same flat parameters draw a structure, fill a calendar, lay out a map, make a molecule, or build a deck. The drawing changes; the link does not.
Every tab is one link. datalink.ac/draw is the same service as x-cmd.com/draw — the address is the only difference. Click any picture to open it in the explorer and change it by hand.
Everything on GitHub is text, and it is read by people and by bots. A datalink goes right in and brings the numbers with it: a reader can check them, a reviewer can change them, a bot can read them, and years later they are still there.
One small skill teaches it to make datalinks properly. No key, no account, no dashboard — the agent composes a link and the picture appears.
npx clawhub@latest install edwinjhlee/x-cmd-draw
curl 'https://datalink.ac/draw?type=pie&labels=TS,Py,Go&values=4821,2312,980'
The skill is optional — an agent that has read this page can already build the links. The prompt in «the flow» above works as it is; the skill just teaches the agent the shape.
A draft specification — the promises a renderer has to keep, and what an agent may rely on. And the reasoning behind it, in the blog.