mnemosyne systems: orangu 1.1.0

mnemosyne systems is happy to announce orangu 1.1.0 - the Open Source AI-powered coding environment.

This is the first feature release since 1.0.0 brought the whole stack - the coding environment, the coordinator, and the native pure-Rust GGUF inference server - together under one project. 1.1.0 keeps that shape and makes it faster, puts the first Metal code on the Mac, and turns the web console into a place where you manage your models rather than only chat with them.

Faster inference

The headline of this release is speed. Attention was implemented separately for each backend, which meant every path had its own performance characteristics and its own bugs. In 1.1.0 attention is unified across the backends, with a single shared implementation behind CPU, Vulkan, CUDA, OpenCL, and ROCm.

Unifying it made it worth optimizing once, properly. Depending on the model, the quantization, and the backend you run, this release measures between 15% and 450% faster than 1.0.0 - the larger gains land where the old per-backend code was weakest.

The release now also ships orangu-bench, so you can measure the difference on your own hardware and your own models instead of taking our word for it.

Metal on the Mac

1.1.0 lands the initial Metal work for Apple Silicon. This is the first step, not the finished backend - it sits alongside the CPU, Vulkan, CUDA, OpenCL, and ROCm backends that have been there since 1.0.0, and it will grow over the next releases. If you run orangu on a Mac, this is the release to start testing with, and the release where your reports are worth the most to us.

The web console grows up

orangu-server serves a browser chat UI from the same binary, with no build step and no external assets. In 1.1.0 it does more than chat:

Model management - list, inspect, and manage the machine’s models straight from the browser. What used to be a command-line-only job is now available in the console
Mermaid support - diagrams in a model’s answer render as diagrams. Architecture sketches, sequence diagrams, and flow charts arrive readable instead of as a block of source
History management - conversations can now be deleted from the console, so the persistent history stays yours to prune

orangu-server also gained a refresh subcommand: when a model you already have has a newer hash upstream, refresh pulls the updated weights rather than making you delete and re-download by hand.

Knowing your machine

orangu reports far more about the machine it is running on in this release. Operating system detection was reworked, and GPU and system information is more accurate and more complete - which matters both when orangu picks a backend for you and when you are filing an issue and we need to know what you are running.

Download statistics were improved as well, so pulling a large model tells you what is actually happening.

Also in this release

word_wrap can now be switched on and off in the configuration - a community contribution
Windows fixes across the terminal interface and the engine
The status bar is quieter, showing Thinking/Working rather than a running commentary
The complete manual ships with the release as PDF and HTML, and is embedded in the binary for offline reading with /manual

47 commits went into this release, and it again carries community contributions. Our thanks to everyone who filed an issue, opened a discussion, or sent a patch.

Upgrading

Configuration files changed in this version. After upgrading, run --init for all binaries:

orangu --init
orangu-coordinator --init
orangu-server --init

The one-liner installer puts the whole stack on the machine in a single step, on Linux, macOS, and Windows:

curl -fsSL https://mnemosyne-systems.github.io/orangu/install.sh | sh

Availability

orangu 1.1.0 is available now for Linux, macOS, and Windows on x86_64 and aarch64, with source and checksums published alongside the binaries. Everything is Open Source under the GNU General Public License v3.0 - the environment, the coordinator, and the inference engine alike.

Commercial support for orangu is available from mnemosyne systems - contact sales .

Be part of the community, and be proud of your contributions ! Try it, star it, fork it, break it, tell us what’s missing.

Thanks for your time !

Overview: https://www.mnemosyne-systems.ai/products/orangu
Repository: https://github.com/mnemosyne-systems/orangu
Release: https://github.com/mnemosyne-systems/orangu/releases/tag/1.1.0