yourai CLI v0.0.4 · yourai.bdtren.com

Changelog

Project-local models.

Models and engines install into yourai_module/ with a root yourai.lock (like node_modules / yarn.lock). Delete the module folder anytime and run yourai build again.
~/.yourai is an optional shared cache: yourai build -s saves there, -f downloads online instead of using it, -f -s refreshes the saved copy.
yourai -v prints the version (same as --version).

Install fixes.

The macOS and Linux installer links yourai into a writable folder already on your PATH, so it works in the same terminal. yourai uninstall removes the link.
Homebrew instructions include the brew trust step that Homebrew 7 requires.
npm publishing skips versions that are already published, so a failed release can be re-run.

Agent and multimodal capabilities.

The wizard asks which capabilities you want and recommends a model for each: agent, vision, images, short videos, speech to text, and text to speech.
yourai chat can read files (text, code, PDF) in a sandboxed workspace, write them after asking, and call the other capabilities as tools. /image and /audio attach files for vision and omni models.
New commands: image, video, speak, and transcribe.
New API endpoints on the same port: /v1/images/generations, /v1/videos/generations, /v1/audio/transcriptions, and /v1/audio/speech.
Image, video, and speech engines load on first use and unload when idle. status, doctor, and update cover them.
13 new models: Qwen2.5-VL, Gemma 3, Qwen2.5-Omni, SD-Turbo, FLUX.1 schnell, Wan2.1, Whisper, and Qwen3-TTS.
Existing yourai.toml files are upgraded automatically, keeping your comments.
The installer explains how to use yourai right away when the new PATH isn't active yet. doctor --network columns line up. License changed to Apache-2.0.

First test release.

Hardware check (doctor) for macOS, Linux, and Windows, with NVIDIA, AMD, and Apple Silicon GPU detection.
Guided project wizard (init) with recommendations, explanations, and a review screen.
12 chat and code models from 0.5B to 70B, with pinned versions and checksums.
Verified, resumable downloads of llama.cpp and models (build).
Background server with browser chat and OpenAI-compatible API; start, stop, pause, resume, status, logs.
Terminal chat with streaming replies.
Safe config editing and smart rebuild.
Local network mode with a generated API key.
update, self-update, uninstall, and shell completions.
Install with curl, PowerShell, Homebrew, or npm.

Ask questions about your own documents, fine-tuning, image editing and image-to-video, Ollama and vLLM runtimes, and starting automatically at login.