What We Mean by Context Packet

If you've used Kontext, you've seen the button: "Send to AI." Click it, and you get a structured export of your canvas — something you can paste into Claude, ChatGPT, or any other model. We call this a Context Packet.

The name is deliberate. It's not a document. It's not a prompt. It's a packet — and the difference matters.

Documents vs. Packets

A document is a container optimized for human reading. It has typography, layout, white space — visual signals that help human eyes navigate content. A 10-page strategy doc has headers so you know where sections begin. It has bullet points so your eye can skim. It has bold text to flag what's important.

AI doesn't need any of that.

When you paste a document into a chat window, the model has to reverse-engineer your intent from formatting that was never meant for it. Headers become arbitrary line breaks. Bullet points become decoration. The visual hierarchy you carefully built? Noise.

A Context Packet is structured for AI consumption from the start:

<node id="goal" type="objective">
  What we are trying to do and why
</node>

<node id="constraints" type="policy">
  What we cannot do, what we are avoiding
</node>

<node id="decisions" type="history">
  What we have already decided, and why
</node>

The AI doesn't have to guess at structure. It's explicit, labeled, and traversable.

Four Layers

A Context Packet carries four types of information:

Data — the actual content. Your ideas, notes, analyses, goals.

Relations — how things connect. This node was derived from that one. This idea extends that one. This is the counterargument to that decision.

Policy — what the AI should and shouldn't do. Constraints, scope, tone.

Protocol — how this should be consumed. Human-readable output? Machine-readable only? Both?

Most prompts only carry the Data layer. The other three are either absent or buried in prose — which means the AI has to infer them, and frequently gets it wrong. Context Packets make all four layers explicit.

The Real Purpose

A Context Packet isn't just a better prompt format. It's a way to make your thinking portable.

When you build a canvas in Kontext, you're not just capturing ideas — you're building an artifact that any AI can immediately understand and act on. You can share it with a colleague using a different tool. You can save it and return to it next month without losing thread. You can pass it to an agent that doesn't have the ability to ask clarifying questions.

The packet is self-contained. It carries its own context.

Documents are written for the person who reads them. Context Packets are written for the AI that executes them. Treating them as the same thing is how we ended up pasting three paragraphs of background into every chat session.

We thought there was a cleaner way. Kontext is what that looks like in practice.