---
license: apache-2.0
library_name: gguf
tags:
  - private-ai
  - sovereign-ai
  - rmdw
  - echols
  - jeshua
  - web-builder
  - code
language:
  - en
---

# Jeshua 0.5

**Jeshua is the build engine behind [Echols](https://echols.ai) — the model that turns one sentence into a finished, premium website.** Where [Jah 1.0](https://huggingface.co/RMDWLLC/Jah-1.0) is the deep private brain that needs serious hardware, Jeshua is the specialist: small, fast, and trained by RMDW to do one thing at an elite level — ship complete, beautiful, working builds.

This is not a demo checkpoint. Jeshua builds the sites and apps that Echols customers ask for every day, in production, right now.

## What Jeshua does

- **One sentence → a complete premium website.** Full multi-section editorial pages: cinematic hero, real photography, services, galleries, testimonials, CTA, footer — fully styled, in a single self-contained file.
- **Real apps too.** Dashboards, tools, calculators, landing pages — working code, not sketches.
- **Fast.** A sparse mixture-of-experts design keeps only ~3B parameters active per token, so it streams builds quickly — and unlike its big brother, **it runs on hardware normal people own**: one consumer GPU, a gaming PC, or a modern Mac.

## The receipts

Before Jeshua took over the Echols build lane, it faced the previous engine — a model **25× its size** — in blind head-to-head judging on real customer-style build requests:

- **24–1** in blind side-by-side picks
- **49/50 (98%)** on the automated build-quality gate the bigger model passed at 68%

Small, trained right, beats big. That was the bet, and Jeshua is the proof.

## Run it yourself

Serve Jeshua with `llama.cpp` on a single GPU or a Mac:

```bash
llama-server --model Jeshua-0.5-Q8_0.gguf \
  -ngl 999 -c 32768 -fa on --jinja \
  --host 0.0.0.0 --port 8080
```

Or with Ollama using the `Modelfile` included in this repo:

```bash
ollama create jeshua -f Modelfile
ollama run jeshua "Build a landing page for a family bakery."
```

**Tip:** builds come out best at `temperature 0.5` with a system prompt that demands full styling and real imagery — the Modelfile ships with RMDW's build persona baked in.

## Why it exists

Big labs sell you a giant generalist and charge accordingly. RMDW trains specialists: Jah 1.0 for deep private chat, **Jeshua 0.5 for building** — each one honest about what it is, each one yours to run. Jeshua is what happens when a small open base is trained hard on one craft by people who ship with it daily.

Want it with zero setup — plus Jah's deep brain, live search, private memory, and one-click deploys to your own GitHub? That's **[echols.ai](https://echols.ai)**, $25/mo.

## The bigger idea

Most AI today is *rented* from a handful of companies. RMDW builds AI you *own* — real models running a real product on hardware we control, with open weights so anyone can do the same. **AI you own, not AI you rent.** → **[rmdw.ai](https://rmdw.ai)**

---

*Jeshua 0.5 — RMDW LLC. The builder. Run it on your terms.*