v0.1.2-alpha

Patch alpha focused on startup reliability and website preview correctness.

Highlights:

• Fixed non-interactive startup failure: MA now checks for TTY/raw-mode support before Ink renders, so piped/non-TTY runs fail with one clear message instead of a half-rendered UI and Ink stack trace.
• Added regression tests for terminal input support detection.
• Reworked the website TUI preview to match the clearer idle-state screenshot with input prompt, shortcuts, and 33k context usage.
• Rebuilt the init provider preview so LM Studio and DeepSeek cards no longer crop long text in the website.
• Bumped package version to 0.1.2-alpha.0.

Verification:

• npm test: 206/206 passing
• npm run package:portable: passed locally
• ma version: 0.1.2-alpha.0
• non-TTY ma invocation now prints a single clear fatal message
• Playwright website screenshots checked for both preview sections
• GitHub Actions release workflow passed and uploaded Linux x64, macOS arm64, and Windows x64 binaries with sha256 files
• GitHub Actions website workflow passed for the published website

Full Changelog: v0.1.1-alpha...v0.1.2-alpha

v0.1.1-alpha

Alpha iteration focused on polish and release readiness.

Highlights:

• Reworked TUI diff rendering with terminal-width-aware truncation using mature ANSI/fullwidth string libraries.
• Added regression coverage for Chinese/emoji README diffs in an 80-column terminal.
• Fixed CLI version reporting to read package metadata instead of a hardcoded string.
• Updated README and website to match the current product positioning: DeepSeek setup in minutes, local Qwen productivity, benchmark-backed small-model optimization.
• Added clearer TUI and init-flow visuals for LM Studio local and DeepSeek official setup.
• Bumped package version to 0.1.1-alpha.0 and published portable bundles for macOS arm64, Linux x64, and Windows x64.

Verification:

• npm test: 204/204 passing locally
• npm run package:portable: passed locally
• ma version: 0.1.1-alpha.0
• GitHub release workflow: macOS, Linux, Windows jobs all passed

Full Changelog: v0.1.0-alpha...v0.1.1-alpha

MA v0.1.0-alpha

MA v0.1.0-alpha is the first public alpha for a local-first coding agent CLI.

Highlights:

• OpenAI-compatible model providers with DeepSeek and LM Studio support.
• /model model switching with remembered defaults and provider/model profile IDs.
• macOS Keychain-backed API key storage.
• Slash command suggestions in the TUI.
• Built-in MCP servers for exec, fs, fs-edit, grep, and web.
• Skills loading from .ma/skills/.
• Session restore, /revert, task archive, and workspace diff display.

Benchmark snapshot:

• Qwen3-30B local model via LM Studio.
• 70 tasks: L0 x10, L1 x30, L2 x30.
• Pass rate: L0 100%, L1 98.7%, L2 95.3%.
• Details are documented in docs/benchmark-results.md.

Portable downloads:

• macOS arm64: ma-0.1.0-alpha.0-macos-arm64.tar.gz
• Linux x64: ma-0.1.0-alpha.0-linux-x64.tar.gz
• Windows x64: ma-0.1.0-alpha.0-windows-x64.zip

Install smoke test:

1. Download the asset for your platform.
2. Extract it.
3. Run ./ma version on macOS/Linux, or ma.cmd version on Windows.
4. Run ./ma init, then ./ma.

This is an alpha release. The core TUI and model flow are ready for early users, but provider coverage, distribution polish, and security UX will continue to move fast.