Onyx Code 2.6.0

Onyx Code now runs more of the newest local models on llama.cpp 0.8.0, and shows a reasoning model's thinking in a panel of its own, apart from its answer. Codebase search finds the values set at the top of a file, a failed start of the AI server says why, and the setup wizard offers a chat model on smaller computers.

Downloading

If you're installing manually:

macOS

• Apple Silicon Mac (M1 or newer): download OnyxCode-darwin-arm64.dmg (smaller, native).
• Intel Mac, or unsure: download OnyxCode-darwin-universal.dmg (runs on both).

Open the .dmg and drag Onyx Code to /Applications/. Both macOS downloads are signed and notarized by Apple, so they pass Gatekeeper and open without security warnings.

Linux (x64)

• Debian / Ubuntu: download OnyxCode-linux-x64.deb and install it with sudo apt install ./OnyxCode-linux-x64.deb.
• Other distributions: download OnyxCode-linux-x64.AppImage, make it executable (chmod +x), and run it. On Ubuntu 24.04 (which lacks libfuse2) run it with --appimage-extract-and-run.

Windows (x64)

Install with winget: winget install OnyxCode (recommended). It runs through Microsoft's App Installer (a trusted Microsoft Store app) with a package listing reviewed in Microsoft's community winget repository, so it verifies the download and gives a smoother first-run experience. The same installer (OnyxCode-win32-x64-winget.exe) is also attached here for manual download; note it is not code-signed yet either way, so Windows may still show an "Unknown publisher" warning. A code-signed installer is planned for a later release.

The *-autoupdate.zip, darwin-*.json, linux-x64.json and win32-x64.json files are plumbing for Onyx Code's
built-in auto-updater. They are not designed for manual install and will give
you a worse first-run experience than the download above.

What's New

Runs more of the newest local models. The bundled AI engine is now llama.cpp 0.8.0 on every platform. Model families released since the last engine load, among them LFM3, DeepSeek V4.1, Muse Spark, K2 Horizon and Maion Coder, and the models you already run answer as they did.

Reasoning models show their thinking apart from their answer. With a model that thinks before it answers (Qwen3, LFM3, DeepSeek R1 and others), on the bundled AI engine, Ollama or an OpenAI-compatible server, the thinking streams into a Thinking panel above the reply while the usual "Thinking..." stays under it. The panel folds when the answer starts and says how long the model thought ("Thought for 6s"), and a click opens it again, in a saved chat too. Only the answer goes back to the model, and only the answer is read for edits and tool calls. If the model runs out of room while thinking, or its thinking goes round in a loop, the reply says so. Conversation summaries, commit message drafts and /init's drafts ask the model to answer directly, where its template or server can be told. /compare shows each model's answer, and its speed counts the thinking.

Codebase search finds the values set at the top of a file. In TypeScript, JavaScript, Python, Go and Rust, top-level constants and a file's opening lines are part of the codebase index, so chat can answer a question about a price threshold, a rate table or a port in a config file from your code.

A failed start of the AI server says why. When the AI engine is missing (an antivirus can remove it), fails its integrity check, or the chat model's file can't be found (deleted, moved, or on a drive that isn't connected), the notice, the model's card in My Models, Restart AI Server and the codebase index's status say which, and what to do.

A chat model for smaller computers. From 3.5 GB of RAM the setup wizard offers Qwen 2.5 Coder 1.5B as the chat model and says what to expect of it; below that it says why no model is listed. Browse Models' "Recommended for Your Hardware" leads with the same model.

Chat and the AI server are steadier. A path with an escaped space (Onyx\ Code.app/notes.txt) is written where it points. Chat keeps a read's line numbers out of the files it writes, however long the line it read. /model, /stats and the context usage's hover say plainly when memory is tight and that 8 bits fit more of the window. The AI server reads your GPU's memory even while a security product scans a newly updated engine. The status bar names your model as AI Ready: <model>.

On Windows, a link to another computer stays a link. A symbolic link in a folder you open that leads to a share on another computer shows in the Explorer, and Onyx Code does not go through it, so opening the project connects nowhere and sends no sign-in, and chat, search and git stay quick. To follow one you trust, add its host to the security.allowedUNCHosts setting and restart. Language servers and other extensions may still follow such a link themselves.

On Linux, the bundled embedding model stays chosen. Choosing it in the Embeddings tab holds from one start of the AppImage to the next, and codebase search keeps working. Moving the app on macOS or Windows keeps it too.

After you update. Your project's codebase index is built again once, to take in the values set at the top of each file. A saved path to an embedding model whose file is gone is cleared at the first start, with the index built again once.

System Requirements

• macOS 11 (Big Sur) or later.
• Apple Silicon (M1 or newer) or an Intel Mac. The Universal download runs natively on both.

Installing

1. Download the .dmg that matches your Mac (see "Downloading" above).
2. Open it and drag Onyx Code into your Applications folder.
3. Launch it from Applications. Because the build is notarized, it opens without a security prompt.

Verification

This build is signed and notarized by Apple. macOS verifies it on first launch, so it opens as an app from an identified developer without security warnings.