Version 01, the latest authored version

Version 01 · the body as written

# The Seed as a Portable Unit of AI Context

I have been thinking about what the smallest useful unit of context should be when Guava Nexus begins interfacing with AI systems through protocols such as MCP.

At first, I imagined that a user's HashName could function as the context address. But that seems too broad. A HashName represents identity and authorship. It tells us who a person is within the network, not necessarily which part of their recorded thought an AI should consume.

I then considered whether an Arrangement of Seeds should become the context object. An Arrangement may be useful for composing larger bodies of knowledge, but it is still constructed from something smaller.

The smallest meaningful unit appears to be the **Seed itself**.

A Seed is already a bounded piece of authored context. It has an identity, an author, Versions, lineage, and relationships to other Seeds. It can therefore potentially function as an independently addressable unit of context that can be passed between humans, applications, and AI systems.

This suggests a simple product hypothesis:

> **Every Guava Seed could become a portable, provenance-aware unit of AI context.**

Today, when someone shares a social-media post, they usually share a URL. Another person opens the URL and reads the content.

A Guava Seed could extend this familiar behavior.

When someone shares a Seed, another person could still open and read it normally. But the Seed could also expose a machine-readable context address that an AI system can resolve.

The same Seed might therefore have two complementary modes of consumption:

**Human:**
Open this Seed and read what was written.

**Machine:**
Use this Seed as context and reason from it.

A Share action on Guava might eventually allow someone to copy an ordinary Seed link or copy an AI-context link. The user receiving that context could provide it to Claude, ChatGPT, Cursor, Grok, or another compatible intelligence system.

For example, someone could send another person a Seed and say:

*Ask your AI what it thinks about this.*

The AI would not merely receive copied text stripped from its source. It could retrieve the canonical Seed together with the contextual information that Guava is authorized to expose: authorship, Version, lineage, evidence, relationships, and potentially other provenance information.

This distinction seems important.

A Seed should not *be* an MCP object in the sense that Guava becomes dependent on MCP. MCP is only one current protocol through which the Seed might be exposed.

The more durable architecture would be:

**Seed = canonical context object**

**MCP = one interface for consuming it**

Other protocols could eventually consume the same Seed without changing what the Seed fundamentally is.

This could make a Seed analogous to a new kind of Internet object.

A webpage is something a browser can retrieve.

A social post is something people can share.

A Git commit is something software development systems can reference.

A Guava Seed could be something that **both humans and artificial intelligences can reference as context while retaining its intellectual provenance.**

The Seed would therefore be more than published content.

It could become a **portable context primitive**.

Its lineage also introduces possibilities that ordinary shared documents do not naturally provide.

An AI could potentially be asked to consume:

**Seed only** — reason from this specific recorded context.

**Seed + ancestors** — understand what this thought grew from.

**Seed + descendants** — examine what subsequently developed from it.

**Seed lineage** — examine the larger intellectual trajectory around the thought.

Arrangements could then exist one level above this system. If a Seed is the atomic context unit, an Arrangement could compose several Seeds into a larger context package without changing the canonical identity or provenance of the individual Seeds.

This creates a possible hierarchy:

**HashName → identifies the author**

**Seed → records the smallest portable contextual object**

**Lineage → records how contextual objects develop from one another**

**Arrangement → intentionally composes multiple contextual objects**

**MCP or another protocol → allows external intelligence to consume them**

The product implication may be significant.

Most AI context today is copied into prompts, hidden inside proprietary memory systems, uploaded as files, or manually reconstructed between applications. The context is frequently separated from the history of how it came into existence.

Guava Nexus could instead make context itself addressable.

Rather than sending someone a paragraph and hoping its surrounding meaning survives, a creator could send the Seed.

The recipient would receive not merely information, but an entry point into its recorded context.

If this works, perhaps the basic social action within Guava changes from:

**Share this post**

to:

**Share this context.**

And eventually:

**Give this thought to your AI.**

## Questions to Explore

This is currently a product hypothesis rather than an implementation decision.

Several questions need exploration:

What exactly must a Seed expose for an external AI to understand it correctly without overloading the context window?

Should the default machine-readable representation contain only the current Version, or should Version history always remain discoverable?

How should lineage traversal work when a Seed is shared as context?

How does an AI distinguish the author's recorded commitments from evidence, citations, branches, contributions by others, and AI-generated synthesis?

Should the same Seed context address work across multiple AI systems and protocols?

What happens when a public Seed later develops substantially through new Versions?

Can permissioned or private Seeds eventually expose revocable context access without compromising provenance?

Should a recipient be able to combine several shared Seed links into an Arrangement and thereby construct a new contextual environment?

Could these portable context objects eventually become one of the primary ways people communicate ideas with AI systems—and with one another?

The central question is therefore no longer merely whether Guava Nexus can store human thought.

It is whether a recorded thought can become a **portable computational object**: something that can move between intelligences without losing where it came from, who committed it, how it changed, and what it grew from.

keccak256 of this body is what the commitment on chain records · the title, the version label, the author's note and the source are outside it