README.md

---
title: Optitransfer
emoji: 🔬
colorFrom: gray
colorTo: gray
sdk: static
pinned: true
tags:
  - model-merging
  - crdt
  - distributed-systems
  - federated-learning
  - convergence
  - deterministic
  - conflict-free
  - byzantine-fault-tolerance
  - agent-memory
  - provenance
  - safetensors
  - pytorch
  - neural-networks
  - multi-agent
---

<div align="center">

# Optitransfer

**Convergent intelligence infrastructure.**

*The mathematical layer where distributed AI systems merge, verify, and evolve -- deterministically.*

---

[![PyPI](https://img.shields.io/pypi/v/crdt-merge?label=PyPI&color=blue)](https://pypi.org/project/crdt-merge/)
[![Downloads](https://img.shields.io/pypi/dm/crdt-merge?label=Downloads&color=brightgreen)](https://pepy.tech/projects/crdt-merge)
[![Tests](https://img.shields.io/badge/tests-4%2C498%20passing-brightgreen)](https://github.com/mgillr/crdt-merge)
[![License](https://img.shields.io/badge/license-BSL%201.1-orange)](https://github.com/mgillr/crdt-merge/blob/main/LICENSE)
[![Patent](https://img.shields.io/badge/UK%20Patent-pending-lightgrey)](https://github.com/mgillr/crdt-merge/blob/main/PATENTS)

</div>

---

## crdt-merge

Every standard merge strategy fails at least one of the three algebraic laws required for distributed convergence. That failure is mathematical, not implementational.

**crdt-merge** is the fix. A patented architecture that makes any merge operation -- across model weights, datasets, or agent memory -- fully conflict-free, order-independent, and provably convergent.

26 strategies. Zero coordination. One correct result.

```
pip install crdt-merge
```

---

## What We Work On

Problems at the intersection of distributed systems theory and machine intelligence. Some of these challenge long-standing assumptions in the field.

- Convergent model merging at arbitrary scale
- Distributed trust and verification in open networks
- Cryptographic provenance for merge decisions
- Knowledge acquisition without catastrophic forgetting
- Coordination without central arbiters

Selected findings are published alongside the library.

---

## Demonstrations

| Space | Description |
|:--|:--|
| [crdt-merge](https://huggingface.co/spaces/Optitransfer/crdt-merge) | Merge real models with full provenance |
| [Convergence Lab](https://huggingface.co/spaces/Optitransfer/convergence-lab) | 26 strategies, convergence visualisation, experiments |
| [Data Playground](https://huggingface.co/spaces/Optitransfer/crdt-merge-data) | CRDT merging on tabular data |
| [Federation](https://huggingface.co/spaces/Optitransfer/crdt-merge-federation) | Multi-node gossip convergence |

---

## Numbers

| Metric | Value |
|:--|--:|
| Merge strategies (CRDT-verified) | 26 |
| Test suite | 4,498 passing |
| CRDT overhead per merge | < 0.5 ms |
| Acceleration vs naive | 38.8x |
| Cross-language ports | Python, Rust, TypeScript, Java |

---

## Links

| | |
|:--|:--|
| Source | [github.com/mgillr/crdt-merge](https://github.com/mgillr/crdt-merge) |
| Paper | [CRDT Merge ArXiv (PDF)](https://github.com/mgillr/crdt-merge/blob/main/paper/CRDT_Merge_ArXiv.pdf) |
| Package | [pypi.org/project/crdt-merge](https://pypi.org/project/crdt-merge/) |
| Rust | [crates.io/crates/crdt-merge](https://crates.io/crates/crdt-merge) |
| TypeScript | [npmjs.com/package/crdt-merge](https://www.npmjs.com/package/crdt-merge) |
| Download Stats | [ClickPy](https://clickpy.clickhouse.com/dashboard/crdt-merge) -- [PePy](https://pepy.tech/projects/crdt-merge) |
| Patent | UK Application GB 2607132.4 |
| License | BSL 1.1 (Apache 2.0 from March 2028) |

---

<div align="center">

Switzerland -- [data@optitransfer.ch](mailto:data@optitransfer.ch)

</div>