Changelog

v1.0.1 - Auto Updates

Released: June 22, 2026

Shared update checking: Both msh and mshell now check npm for new versions on startup and show an update banner when one is available.
Configurable auto-update: Opt in with msh --auto-update or /auto-update in the TUI to install updates automatically (uses bun or npm).
Manual update command: Run msh --check-update or /update in the TUI to check and install on demand.

v1.0.0 - Unified TUI UX

Released: June 22, 2026

Magic Shell 1.0 marks the first stable release of the refreshed terminal experience.

Unified command palette, slash command menu, model picker, provider picker, theme picker, onboarding provider picker, safety picker, thinking picker, and API-key setup around a consistent modal style.
Slash commands now run the selected action directly, so /models, /providers, /themes, /safety, and /thinking open the proper picker instead of submitting text as a normal prompt.
Added explicit safety and thinking level pickers. Ctrl+Y opens safety levels and Ctrl+T opens thinking levels.
Improved modal cancellation: Esc and Ctrl+C close the active modal before exiting the TUI.
Improved model and provider descriptions, cost labels, and footer helper text so picker context is easier to read.
Bumped package version to 1.0.0.

v0.2.24 - Cost-Tier System

Released: June 22, 2026

What’s New

Cost tiers on every model: Replaced the ambiguous free?: boolean flag on Model with an explicit cost: CostTier field — "free" | "lower-cost" | "premium". The CLI and TUI previously had no way to distinguish cheaper paid models from expensive premium ones, so users could unknowingly select GPT 5.5 Pro at $30/$180 per 1M tokens for a one-line command translation
Cost-aware model picker: The TUI model selector (showModelSelector) and CLI msh models now sort models by cost tier (free → lower-cost → premium), then by category (fast → smart → reasoning), then alphabetically. Each entry shows a [FREE] / [lower-cost] / [premium] label so users can pick with intent
Highest-cost models removed: GPT 5.5 Pro (the only premium tier model in the registry) was removed from every provider — $30/$180 per 1M tokens is not appropriate for terminal command translation
sortModelsByCost() helper: New exported helper in src/lib/models.ts centralizes the cost-tier sort order used by both the CLI and TUI

Improvements

Generated model catalog: website/scripts/generate-models-data.ts now emits the cost field instead of free, and ModelCatalog.astro reads cost directly rather than inferring it from free + category. Catalog filters and badges now reflect the source-of-truth tier
AGENTS.md type docs updated: The Model interface example now shows the cost: CostTier field and the new CostTier type

v0.2.23 - Docs Theme, Provider Cards, and Model Catalog

Released: June 22, 2026

What’s New

Slimmed model registry: Reduced the built-in model list to a curated 6-8 per provider — the latest available versions spanning free, lower-cost, and premium across fast, smart, and reasoning categories. Removes 30+ stale entries that had drifted from upstream offerings
Free model on every major provider: OpenCode Zen ships deepseek-v4-flash-free and OpenRouter now ships deepseek/deepseek-v4-flash:free — both free, so users can try Magic Shell without a paid API key on either provider
Model catalog page: New /reference/models/ page lists every supported model across OpenCode Zen, OpenRouter, Vercel AI Gateway, Cloudflare AI Gateway, and Workers AI — generated from src/lib/models.ts at build time, so the docs can’t drift from the CLI
Filterable catalog: Provider, category, and cost filters on the catalog page; cost legend reuses the existing Free / Lower-cost / Premium vocabulary
Provider setup cards: Replaced the cramped six-tab “Get an API Key” block on the Configuration page with an accessible card grid using native <details>/<summary> — works without JS, keyboard-operable, and multiple providers can be expanded at once

Improvements

Theme switcher fixed: Light and dark palettes now live under explicit [data-theme] selectors, so toggling the Starlight theme switcher actually changes colors. Light mode ships a warm off-white palette tuned to the same orange accent
Aside and blockquote readability: Beefed up callout borders to 4px with rounded corners; blockquotes now have a tinted background and 3px bar in both modes
Providers page cleanup: features/providers.mdx no longer carries hand-maintained per-provider model tables (which had drifted from source); it now links to the generated catalog. The curated “Model Equivalents Across Providers” table is retained

Fixes

Corrected gpt-5-nano free-flag on OpenCode Zen — it is a paid lower-cost model ($0.05/$0.40 per 1M tokens), not free. The only true free Zen models are deepseek-v4-flash-free, mimo-v2.5-free, north-mini-code-free, nemotron-3-ultra-free, and big-pickle

v0.2.22 - Model Catalog and Release Prep

Released: June 16, 2026

What’s New

Refreshed OpenCode Zen with the current free model set, including deepseek-v4-flash-free, mimo-v2.5-free, north-mini-code-free, nemotron-3-ultra-free, big-pickle, and gpt-5-nano
Added current OpenRouter open-source/open-weight highlights, including moonshotai/kimi-k2.7-code, z-ai/glm-5.2, minimax/minimax-m3, qwen/qwen3.7-max, and qwen/qwen3.7-plus
Updated OpenCode Zen Qwen model routing to use the Messages endpoint documented by Zen

Improvements

Updated the default model to deepseek-v4-flash-free
Refreshed release-facing README, provider docs, config docs, and FAQ schema copy
Updated website dependency lockfiles for release prep

v0.2.20 - Thinking Controls and Zen Free Models

Released: May 1, 2026

What’s New

Added --thinking <level> and thinkingLevel config support for off, low, medium, and high
Added provider-specific reasoning controls for supported OpenRouter, OpenCode Zen, gateway, and custom models
Updated OpenCode Zen free models, including minimax-m2.5-free, ling-2.6-flash-free, hy3-preview-free, nemotron-3-super-free, trinity-large-preview-free, big-pickle, and gpt-5-nano

Improvements

Disabled custom thinking parameters when the selected model or provider does not support them
Updated default model references to minimax-m2.5-free
Documented thinking controls in CLI options and config reference docs

v0.2.19 - AI Gateway Providers

Released: May 1, 2026

What’s New

Added Vercel AI Gateway, Cloudflare AI Gateway, and Cloudflare Workers AI as first-class providers
Added provider setup, provider switching, model listing, and API-key handling for the new gateway providers
Added gateway environment variable support for AI_GATEWAY_API_KEY, CLOUDFLARE_API_TOKEN, CLOUDFLARE_ACCOUNT_ID, and CLOUDFLARE_AI_GATEWAY_API_KEY

Improvements

Refreshed built-in model registries with current OpenCode Zen and OpenRouter model IDs
Added latest OpenAI and Anthropic aliases/models across OpenRouter and Vercel AI Gateway
Added provider display-name and provider-model helpers so model selection is scoped to the active provider

v0.2.18 - Model Registry Refresh

Released: May 1, 2026

Improvements

Added latest MiMo V2.5 and MiMo V2.5 Pro entries for OpenCode Zen
Added latest Xiaomi MiMo V2.5 and MiMo V2.5 Pro entries for OpenRouter
Removed the older OpenRouter MiMo V2 Flash free entry from docs and model listings
Updated runtime dependencies for the published CLI package

v0.2.17 - Dependency Maintenance

Released: April 2, 2026

Improvements

Updated the CLI package dependencies to current releases
Fixed TUI input typing compatibility after the @opentui/core upgrade

v0.2.15 - Slash Command Support

Released: January 30, 2026

What’s New

Slash Commands: You can now use / prefix for TUI commands (for example /help, /clear, and /models). Both ! and / prefixes work.
Updated help text in TUI to show available commands with both prefixes
Updated documentation to reflect the new command options

Improvements

More intuitive command prefix options for users coming from other CLI tools
Consistent command behavior regardless of prefix choice

v0.2.14

Released: January 23, 2026

Improvements

Improved CI/CD workflows with better automation
Enhanced landing page SEO and meta tags
Fixed bin paths in package.json for better cross-platform compatibility

v0.2.13

Released: January 22, 2026

Major Features

Custom Model Support: Add your own local or remote OpenAI-compatible models (LM Studio, Ollama, etc.)
Secure Credential Storage: API keys are securely stored in your system keychain (macOS Keychain, Linux secret-tool, Windows Credential Manager)
Project Context Detection: Optional detection of package.json scripts, Makefile targets, and more for smarter translations
Multiple AI Providers: Support for OpenCode Zen (with free models!) and OpenRouter
Beautiful TUI: Full-featured terminal interface with 8 built-in themes
Comprehensive Safety System: Multi-level safety analysis with severity levels (critical, high, medium, low)
Command History: Context-aware translations based on recent commands
Auto Updates: Automatic update checking with one-command upgrades

Initial Release

Magic Shell transforms natural language into terminal commands with AI-powered translation and built-in safety features.

Version History

Version	Date	Highlights
v1.0.1	Jun 22, 2026	Auto updates
v1.0.0	Jun 22, 2026	Unified TUI UX
v0.2.24	Jun 22, 2026	Cost-tier system
v0.2.23	Jun 22, 2026	Docs theme, provider cards, model catalog
v0.2.22	Jun 16, 2026	Model catalog and release prep
v0.2.20	May 1, 2026	Thinking controls and Zen free models
v0.2.19	May 1, 2026	AI Gateway providers
v0.2.18	May 1, 2026	Model registry refresh
v0.2.17	Apr 2, 2026	Dependency maintenance
v0.2.15	Jan 30, 2026	Slash command support
v0.2.14	Jan 23, 2026	CI/CD improvements
v0.2.13	Jan 22, 2026	Initial public release