File size: 3,318 Bytes
17276bf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 | ---
license: mit
library_name: transformers
pipeline_tag: text-generation
tags:
- private-ai
- sovereign-ai
- rmdw
- kiyomi
- jah
- fp4
- fp8
- blackwell
- vllm
language:
- en
---
# Jah 3.0
**Jah is the private AI that powers [Kiyomi](https://kiyomibot.ai), RMDW's private alternative to ChatGPT and Claude.** It runs entirely on hardware RMDW owns and controls. Nothing you type leaves to a third-party cloud, nothing is stored externally, and nothing is ever used to train another company's model. What you bring to Jah stays yours.
This is not a chatbot demo. Jah is the brain of a full private-AI product that people pay for and use every day, and this repo holds the exact weights serving them right now. Open, downloadable, verifiable.
## What's new in 3.0
Jah 3.0 is the fastest and longest-context engine the family has ever run, and the first built for native tool use end to end.
- **Four times the speed.** Around 276 tokens per second on RMDW hardware, up from ~65 in Jah 2.0, with the first token back in about 40 milliseconds.
- **Four times the context.** A 1,048,576 token window, up from 250k. A whole repository with its history, or a full document set, held in one conversation. Verified, not assumed: a fact planted deep inside a 380,000-token prompt is retrieved correctly in about 50 seconds.
- **Native tool calling, everywhere.** Files, terminal, browser, macOS control, connectors, web search, and sub-agents all run through the model's own function-calling format. On our own benchmark, moving to native tools was worth 45 points of task completion.
- **Verify before done.** Jah is not allowed to call a coding job finished if it changed code and never ran it. The harness holds the run open until the work is proven. A model that checks its work beats a faster one that ships blind.
- **Held to a frozen bar.** Thirty-two fixed agentic tasks run in throwaway sandboxes with real execution. No engine ships unless it holds the floor. This gate has already caught and rejected a candidate that regressed.
## Measured against Jah 2.0
The same frozen benchmark, the same harness, the same scoring. "Done" means the run finished cleanly; "verified" means it wrote code and then ran something to prove the code worked.
| | Standard set | Hard set | Avg per task |
|---|---|---|---|
| Jah 2.0 | 100% done / 85% verified | 100% / 100% | 20s and 81s |
| **Jah 3.0** | **100% done / 90% verified** | **100% / 100%** | **5s and 12s** |
Higher verification, and roughly seven times faster, on hardware RMDW owns.
## What Jah does in production
- **Private chat** with artifacts, persistent memory, cited web search, and code execution.
- **Builds real apps and sites.** Describe it; Jah writes the project, pushes it to your own GitHub, and deploys it to a live URL you own.
- **Runs the developer API.** Point Claude Code, Codex, or any OpenAI/Anthropic-compatible tool at kiyomibot.ai and Jah writes and refactors code on private hardware.
- **Runs on your desk.** The Kiyomi macOS app drives files, terminal, and the browser as a full agentic coder, and an iPhone remote controls it from anywhere.
All of it private. Try it at **[kiyomibot.ai](https://kiyomibot.ai)**.
---
*Built by [RMDW AI Consulting](https://rmdw.ai). Private AI on hardware you can point at.*
|