🌍 Universal AI Clipboard (UAC) — A New Category in AI Agent Architecture. Call to the community to implement.

Universal AI Clipboard (UAC)

What Is This? (Simple Explanation)

AI agents waste up to 90% of their output by constantly regenerating the same code, commands, and text over and over. Every time an AI writes a command, a code block, or a configuration — even if it just wrote the exact same thing a minute ago — it generates it from scratch.

Universal AI Clipboard (UAC) fixes this by giving AI agents the ability to copy and paste between their own tool calls, just like humans use Ctrl+C / Ctrl+V.

Instead of: "write this long command from scratch"
The agent says: "take command from previous step and add these flags"

Result: up to 96% fewer tokens wasted, 60-80% cheaper AI sessions, 5x less energy consumption.

What Makes It Unique?

All existing AI citation systems (Anthropic Citations, OpenAI Citations, Google Grounding) work on Agent to User axis — they show the user where the AI got information.

UAC works on the Agent to Agent axis — it allows the AI to reuse content between its own tool calls. This is a fundamentally different category.

CO-DISCOVERER: INDEPENDENT DISCOVERY

I acknowledge and respectfully recognize that TJ Guadagno (TJ-Codes) independently arrived at the same fundamental idea and experimentally implemented it as Clipboard Primitives — a copy/template_invoke mechanism with named slots and placeholder syntax for AI agents.

His work: Copy and Paste for AI Agents: An Experimental Primitive (~December 2025)
Code: github.com/TJ-Codes/Agent-clipboard

TJ Guadagno is an independent co-discoverer of this concept. His experiment confirmed that:

Agents naturally adopt clipboard semantics without being taught
Content flows byte-for-byte identical (zero mutation risk)
Measurable token savings (38%+)

My Universal AI Clipboard concept is broader and more universal — it includes Clipboard Primitives as a special case and extends the idea to:

5 Core Mechanisms (Syntactic Clipboard, Anchor Pair, Message Index Map, Transform Pipeline, Cross-Protocol Citation)
Mosaic Assembly (combining fragments from multiple sources)
Clipboard Manager (named slots with persistent storage)
Inline Injection into any tool parameter including MCP

Computer analogy: Clipboard Primitives is like a simple copy utility for one application. Universal AI Clipboard is a full system-wide clipboard manager working across all applications, with slot management, transform pipeline, and cross-protocol integration.

5 Core Mechanisms

#	Mechanism	Benefit
1	Syntactic Clipboard — reference any function by name	99.7% token savings
2	Anchor Pair Citation — cite by start/end markers	80-96% savings
3	Message Index Map — precise citation from chat history	~95% savings
4	Transform Pipeline — edit fragments on the fly	All in 1 call
5	Cross-Protocol Citation — embed in any tool	Zero schema changes

Mosaic Assembly and Clipboard Manager

The agent becomes an editor and curator — combining fragments from different sources (files, chat, terminal, APIs), editing on the fly, reordering, nesting, and building composite documents. With named storage slots, the agent can save, reference, and reorganize content across sessions — all at zero token cost for storage operations.

Global Impact

Up to 96% token savings, 60-80% cost reduction, billions of tokens saved daily, tens of thousands of tons of CO2 reduced annually.

TIMELINE

Date	Event
~2021	First awareness of the problem (DScoNOIZ): thought "model should not regenerate what it has already seen"
~December 2025	Independent experiment: TJ Guadagno publishes Clipboard Primitives with copy/template_invoke and named slots
May - June 2026	Research: analysis of existing AI agent architectures
June 2026	Discovery: no existing technology has a complete analogue
June 3, 2026	This manifesto published - discovery released into the public domain
June 4, 2026	TJ Guadagno's work discovered - acknowledgment of independent co-discoverer

Current Status

I am currently trying to implement this technology in Zoo-Code through vibe coding in my local repository. I am not a programmer and do not understand code — I am doing this exclusively with the help of AI agents.

The implementation in Zoo-Code is called Content Reference (CRT) mechanism — already partially implemented with ref, multi_ref, transform parameters in the codebase.

Call to Action

But I believe this technology should be shared with the world right now, without waiting for a complete implementation. We need to:

Save billions of tokens wasted on regeneration
Reduce energy consumption of the AI sector
Decrease carbon emissions - tens of thousands of tons of CO2 per year
Make AI agents 5-10x more efficient

I call upon the community:

Do not wait - start implementing UAC in your own stacks
Add the Content Reference mechanism to your frameworks and tools
Share this discovery with everyone who works with AI agents
Help make this a global standard

Resources

Full Manifesto: https://dsconoiz.github.io/UNIVERSAL-AI-CLIPBOARD
Co-Discoverer's Implementation (TJ-Codes): https://github.com/TJ-Codes/Agent-clipboard
Co-Discoverer's Article: https://www.linkedin.com/pulse/copy-paste-ai-agents-experimental-primitive-tj-guadagno-v3r9c
License: CC0 1.0 Universal - Public Domain (no patents, no royalties, no restrictions)

Let AI agents stop rewriting the same thing over and over. Let them start assembling mosaics.

This manifesto may be freely translated into any language.

@DScoNOIZ

@DScoNOIZ The clipboard idea is something you can implement as an MCP server and a SKILL combination. You let the MCP server have 2 tools: "clipboard_add" and "clipboard_read". It's simple, in-memory storage; doesn't even need a DB. Just Array of strings (stack). You can set a max entries list (context limit/management). And the tool description would be something like: "Add any correctly working CLI/shell command to remember it precisely"; and for clipboard_read: "Returns any correctly working CLI/shell command you want to remember". And then you add a SKILL instructing for always checking on using these tools before calling CLI/shell. This should be quite simple to even vibe code as a quick demo.

Actually this is something like an episodic agent memory, just simpler and SHORT TERM.

You might be interested in the long-term solution too: https://github.com/TencentCloud/TencentDB-Agent-Memory/tree/main

@kyr0

@kyr0 Thank you so much for showing interest! Honestly, I’m quite surprised by the rather weak response from such a highly respected community. You are the first person who took the time to write me your own constructive vision—and specifically on GitHub. I’m truly touched. For some reason, most people remain silent.

Regarding what you described, yes, to some extent you are right. However, based on my current analysis and conversations with artificial intelligence, I have concluded that this can only be implemented through close integration with each specific platform, taking into account all the nuances.

So, as you mentioned, a partial implementation is indeed possible—but it won’t be able to fully intercept and leverage all the capabilities of a given platform. What is required is deep integration at the level of tools and code.

Our goal is to create a universal clipboard, a copy-paste system for AI. And an MCP server is, roughly speaking, an isolated system in itself. Therefore, our clipboard needs to work with MCP servers as well—all connected servers included—so that, when necessary, we can copy from them and paste back into them the information we need.

In other words, we need a way to extract information from them via copy-paste and, for example, insert it into a chat, a file, or any other destination. At the same time, we should be able to send information into MCP servers via copy-paste. This is essentially how I envision it.

Navigation Menu

Repository navigation

🌍 Universal AI Clipboard (UAC) — A New Category in AI Agent Architecture. Call to the community to implement. #464