⏺ Discovery Manifesto
I announce the discovery of a fundamentally new category in AI agent architecture — the "Optional Inter-Tool Content Citation Mechanism", named Universal AI Clipboard (UAC).
An AI agent can reference existing contextual content through dedicated parameters (ref, multi_ref, transform) instead of regenerating it. The model becomes an assembler, combining ready-made information fragments in a single tool call.
I release this discovery into the free use of the global community under CC0 license — no patents, royalties, or restrictions.
⏺ The Problem: Regeneration Crisis
Modern AI agents waste up to 90% of output traffic regenerating content that already exists in the session context.
Millions of AI agents daily regenerate terabytes of existing content, consuming gigawatt-hours of energy. This costs billions of dollars and produces tens of thousands of tons of CO₂ annually.
⏺ Five Core Mechanisms
🔗 Syntactic Clipboard
The model specifies a single anchor — a function name, class name, or variable. The system automatically finds it and extracts the entire syntactic unit.
📇 Anchor Pair Citation
The model specifies only the START and END of a fragment. The system finds everything in between using multi-stage matching, even tolerating minor typos.
🧠 Message Index Map
A separate index maps exact character positions in chat history. The model writes a short reference like "167:14..18" — the system extracts with 100% precision.
🔄 Transform Pipeline
A copied fragment can be modified on the fly — replace, prepend, wrap, append, or merge multiple fragments. All in a single tool call.
🔁 Cross-Protocol Citation
Reference markers can be embedded inside any tool's text parameters. The system substitutes real content before calling external servers.
🏖️ Isolated Sub-Sessions
For complex tasks, a helper agent runs in an isolated temporary session, performs dirty work, returns only the clean result, then is destroyed.
⏺ Uniqueness Verification
| Mechanism | Axis | Purpose | UAC coverage |
|---|---|---|---|
| Clipboard Primitives (TJ Guadagno) | Agent → Agent | copy+template_invoke, named slots, {{slot}} placeholders, harness-level resolution, byte-identical transfer | ⚠️ ~40% |
| Anthropic Citations API | Agent → User | Show source of answer | ❌ |
| OpenAI Citation Formatting | Agent → User | Tags for source display | ❌ |
| Google Gemini Grounding | Agent → User | Web search confirmation | ❌ |
| ContextCite (MIT CSAIL) | Agent → User | Source identification | ❌ |
| ★ Universal AI Clipboard (UAC) | Agent → Agent | 5 mechanisms, all sources, transforms, MCP injection, mosaic assembly | ✅ 100% |
Conclusion: Universal AI Clipboard (UAC) has no complete analogues in worldwide AI agent systems practice.
⏺ Deep Research Verification (10 Rounds)
I conducted 10 rounds of deep search (Tavily advanced depth, ~100 sources) across all known AI categories: arXiv, LangChain, Anthropic, OpenAI, Google, MCP, CAMEL-AI, UiPath, Microsoft, AWS, GitHub discussions — zero complete analogues found.
| Category | Sources | Analogue found? |
|---|---|---|
| Human clipboard tools | UiPath Clipboard AI, PowerToys Advanced Paste | ❌ For humans, not AI agents |
| Tool output caching | CAMEL-AI "Brainwash Your Agent" | ⚠️ Only 1 last output, no 5 mechanisms |
| Academic research | AgentReuse (arXiv), KVCOMM | ❌ Plan/KV reuse only, not content |
| LangChain artifacts | content_and_artifact, ToolMessage.artifact | ❌ Content/metadata split, not citation |
| MCP protocol | MCP Resources, MCP Tools | ❌ Tool connection standard, not citation |
| Citation APIs | Anthropic/OpenAI/Google/Azure | ❌ Agent→User axis only |
| Shared memory | Multi-agent memory systems | ❌ RAG memory, not precise citation |
| ★ UAC (this discovery) | Universal AI Clipboard | ✅ 5 unique mechanisms, new category |
⏺ Resources
Full Manifesto (English)
Complete description: problem, solution, 5 core mechanisms, economic impact, call to action.
Полный манифест (русский)
Всеобъемлющее описание: проблема, решение, 5 механизмов, экономический эффект, призыв к действию.
GitHub: UNIVERSAL-AI-CLIPBOARD
Source documents, translations, all research materials and proof of discovery date.
Full Research History
10 rounds of deep search across ~100 sources. Timeline of all discoveries including co-discoverer.
Comparison With Existing Systems
Detailed feature-by-feature comparison: UAC vs IBM Memory Pointer, AWS Strands, LangChain, Claude Code, CAMEL-AI, Clipboard Primitives.
Technical Specification
JSON schema, resolution algorithms, transform pipeline specification, and refutation of common objections.
TJ Guadagno — Clipboard Primitives
Independent co-discoverer. Experimentally implemented clipboard primitives for AI agents (~Dec 2025): copy + template_invoke with named slots. Proved models adopt clipboard semantics naturally.
I Call Upon the Global Community
AI framework developers, agent creators, researchers, business, environmentalists — this technology belongs to everyone. Fork, implement, improve. Released into the public domain.
📖 Read the Manifesto ⭐ GitHub