aituner v0.3.0

aituner 0.3.0

New

• Stop all. One control at the top (the phone header and the menu bar panel too) stops every model-related thing
  in order and gracefully: chat answers (saved), the coding agent, a running benchmark, knowledge indexing, the
  gateway, the model server, Ollama, leftovers from aituner's MLX environment, and monitoring. Chat and Coding stay
  locked until you restart from the same dialog, with or without the models that were running.
• New coding projects. Coding creates a project in ~/Projects as a git repository on main with a first commit,
  and has a folder browser to open existing ones.
• Remove everywhere. Every downloaded model has Remove (Setup, Discover, On this Mac). opencode, git and GitHub CLI
  are in Tools and services with Stop or Uninstall; Bootstrap installs git and GitHub CLI.
• Locked sections say why. Chat and Coding unlock once a model runs; every locked section explains itself.
• The menu bar panel opens for a moment when a model starts, as a sign it is loading.
• Knowledge base groundwork. The index and hybrid search for your knowledge folder are in place; Chat will use
  them in a later release.

Fixed

• The Coding terminal is always black, so opencode draws its dark theme correctly (it drew light panels in dark mode).
• Paths in Coding show as ~/... for any home folder.

Other

• The disk image is now named aituner_v0.3.0.dmg.
• The website's screenshots and docs are refreshed with every release; new page: Stop all.

Signed with Developer ID and notarized by Apple. Verify with SHA256SUMS; spctl --assess -v /Applications/aituner.app
should say "accepted".

aituner v0.2.0

aituner 0.2.0

Chat and a coding agent on your local model, a download queue you can always see, and release notes before every update.

Chat

• Chat with the running model under Monitor: saved sessions, answers that stream in, Stop that keeps what was written.
• Code blocks with copy, tables and lists; a reasoning model's thinking folds away.
• Answer again, or edit a message and send it again. Export a chat as Markdown.
• Memory you control: Remember on any message, or add, edit and delete items. Nothing is remembered on its own.
• Every model setting in a side panel (system prompt, temperature, top P, min P, top K, length, seed, thinking, penalties, stop sequences, history), with Reset to defaults: the model authors' recommended values when the model has them.

Coding

• A Coding page runs opencode, an open-source coding agent, on your local model: file tree, an editor with Compare with HEAD, and the agent in a real terminal.
• It asks before it edits a file or runs a command. Only the local model is enabled; sharing and web search are off; your own opencode settings are untouched.
• The agent's edits appear in the editor as they happen; changed files are marked.
• Install it from the page with one confirmation (Homebrew).

Models and downloads

• Queue has its own place in the sidebar with a progress ring; the menu bar panel shows the current download.
• Added models show at once, and Continue to Setup stays in reach once one is queued or ready.
• Remove a model from On this Mac.
• Unsupported Models lists what canirun.ai ranks but aituner cannot run, and why.

Monitor

• Bootstrap installs all three terminal monitors (macmon, mactop, nvtop); each can be uninstalled from Monitor.

Updates

• The update prompt shows the release notes; What's new in the banner opens them in aituner, and About shows them.

Fixes

• Some models ended every answer with a visible end-of-turn marker (<|im_end|>). It no longer reaches Chat, editors or OpenAI-compatible tools.
• On phones, the settings fields no longer zoom the page when tapped.

Requires macOS 14 or later on Apple Silicon.
Install: brew install --cask tonynv/tap/aituner or download from https://aituner.app/download/
Contact: info@aituner.app

aituner v0.1.0

aituner 0.1.0

The first release: a native Mac app for running local AI models on Apple Silicon.

• Start-up scan: every launch detects the chip, CPU and GPU cores, memory, storage, power, thermals and runtimes, live.
• Models: canirun.ai-ranked, MLX builds that fit this Mac's GPU memory budget, with a background download queue
  (weights, config and tokenizer only, verified and resumable).
• Setup: Bootstrap creates the models, reports and knowledge base folders and installs MLX and macmon after one
  confirmation; start a model and connect Claude Code, VS Code, Neovim, Vim + tmux or any OpenAI-compatible tool
  through a local gateway (OpenAI and Anthropic Messages APIs).
• Monitor: live GPU use and frequency, CPU, power, temperatures and memory; a menu bar panel with the live model and
  GPU; closing the window keeps aituner and the model running.
• Optional benchmark, tuning (only changes you approve, each reversible) and before/after reports.
• Storage: choose where models, reports and the knowledge base live; clear what aituner created, with a preview.
• Tools and services: see what is running (MLX, model server, gateway, macmon, Ollama) and quit or remove it.
• Updates: aituner checks GitHub for new releases, asks first, verifies the Developer ID signature, notarization and
  checksums, then restarts. Homebrew installs update with brew.
• Dark by default, with a light theme one click away.
• Signed with a Developer ID and notarized by Apple.
• Private by design: listens on 127.0.0.1 only; talks to canirun.ai and Hugging Face for model data and GitHub for
  updates, nothing else. No account, no telemetry.

Requires macOS 14 or later on Apple Silicon.
Install: brew install --cask tonynv/tap/aituner or download the disk image from https://aituner.app/download/
Contact: info@aituner.app