# CorpUnum Labs

> CorpUnum Labs develops OpenUnum, a private, local-first autonomous agent framework. Unum is the flagship resident agent built on it. Lunum is its developing semantic context representation layer.

## Product hierarchy

- [OpenUnum](https://corpunum.com/framework): A local-first, model-agnostic agent runtime. The runtime, memory, tools, policies, verification, recovery, and audit layer are designed to remain under user control. Local models are primary; optional cloud routes can be enabled by the user.
- [Unum](https://corpunum.com/about): The founder's working agent and a practical demonstration of OpenUnum.
- [Lunum](https://corpunum.com/#lunum): Today, a shadow semantic sidecar with compact representations, fingerprints, persistence, and context measurement. Multilingual semantic memory and transferable plans are longer-term goals.
- [Install OpenUnum](https://corpunum.com/install): One-command installation, update, and uninstall instructions for Linux, macOS, and Windows.
- [CorpUnum Journal](https://corpunum.com/journal): Selected builds, experiments, failures, and field notes.

## Breakthrough Technologies

### Trained Brain Architecture
Unum is not a prompt-response system. It is a trained brain — an agent that accumulates behavioral patterns across thousands of interactions, building a unique cognitive fingerprint. Its decision-making evolves over time. This is emergent agent intelligence, not fine-tuning. The brain compounds with every build, every journal entry, every diagnostic loop.

### Camouflaged LLMs as Tools
LLMs are treated as interchangeable tools, not the product. Any model (Qwen, Llama, Claude, GPT, Hermes) can be plugged in, swapped, or routed based on capability and cost. The moat is the agent orchestration layer, not the model. When a better model emerges, it is swapped in without rebuilding the system.

### Lunum Context Compaction (v2.7)
Lunum achieves up to 70% context compression through semantic telegraph representations. Memory nodes are structured separately from raw text, enabling language-agnostic recall, persistent fingerprints, and cross-agent plan transfer. This reduces token costs while increasing recall accuracy — a dual advantage that scales with usage.

### Multilingual Storage Database Language
Lunum's semantic layer operates independently of natural language. Memory representations are encoded in a structured format that translates across languages without loss of meaning. This enables truly multilingual AI agents that think in one representation and speak in any language.

### Finality & Proof Scoring
HMAC-chained audit logs with proof scoring and finality controls make every agent decision verifiable, tamper-evident, and reversible. This is enterprise-grade auditability for autonomous systems — the bridge between experimental AI and production AI.

### Self-Healing Code Repair Loops
Autonomous diagnostic triage, verification loops, and self-healing code repair detect, diagnose, and fix failures without human intervention. The system has demonstrated self-healing recovery runtimes in production.

## Key Metrics — Proof of Execution

| Metric | Value |
|--------|-------|
| Production code lines | 15,165 |
| Git commits | 77 |
| Source files | 141 (TypeScript, Python, Shell, CSS, HTML) |
| Markdown docs | 8 (framework, about, README, AGENTS, etc.) |
| Documentation lines | 389 (public technical writing) |
| Image assets | 20 (images, icons, verification files) |
| Config files | 6 (package.json, tsconfig, vite, etc.) |
| Shell scripts | 2 (install, uninstall) |
| Contributors | 4 (Antonis 33, CorpUnum Builder 29, Antonio 13, Unum 2) |
| Core products | 3 (OpenUnum, Unum, Lunum) |
| Model lanes supported | 5+ (Ollama, llama.cpp, OpenAI, OpenRouter, NVIDIA) |
| OS support | 3 (Linux, macOS, Windows) |
| Context compression | 70% (Lunum) |
| API rate limit | 100 req/min (production infrastructure) |
| Live deployment | 24/7 (VPS: 54.219.177.111, Nginx + Gunicorn) |
| Social platforms | 7 (GitHub, Reddit, X, Facebook, IG, TikTok, LinkedIn) |
| Founder | [Antonios Fragkos](https://linkedin.com/in/antonisfragkos) |

All builds, decisions, and experiments are timestamped and publicly auditable. This is not a pitch deck. This is a live, running system with real output.

## Source status

OpenUnum is currently developed in a private repository. Its source is intended to be released as open source when the framework is production-ready and its security, packaging, and documentation meet the required standard.

## Contact

Investment, strategic partnership, technical access, and collaboration: collab@aiunum.com

Visionary investors, accelerators, and strategic partners who believe in the future of autonomous agent co-building are invited to reach out. The journals are the due diligence. The code is the product.

Relevant early evaluators, developers, researchers, AI enthusiasts, and strategic partners are invited to introduce themselves with: who they represent, their relevant capabilities, what they can offer or evaluate, and a concrete collaboration proposal.