v0.3.0

Added

• npm registry distribution. many-ai-cli is now installable with
  pnpm add -g many-ai-cli (fallbacks: bun install -g / npm install -g).
  Each platform's native Go binary ships in an optional dependency package
  (many-ai-cli-<os>-<arch>), so no standalone exe is downloaded in a browser
  and no Mark-of-the-Web SmartScreen prompt is triggered. The release workflow
  stages the GoReleaser binaries into the npm packages and publishes them with
  provenance.
• Windows release archives now include unblock-windows.cmd, a local helper
  that runs PowerShell Unblock-File for many-ai-cli*.exe files after zip
  extraction.
• Detached Session Grid. AI and Shell sessions can be popped out into a
  separate browser window as a standalone grid (with presets such as
  "Claude + Shell 2x2" / "Shell 3x3"); the Hub keeps managing approvals and
  session state from the main window.
• Shell sessions. A plain interactive shell (PowerShell / bash / sh) can be
  spawned as a regular Hub session alongside AI sessions; AI-only features
  (approval injection, Chat, token bar) are disabled automatically for shell
  sessions.
• Always-on status bar. A single bottom line shows the active session's
  state, provider/model, work label, project/branch and changed-file count,
  context-window gauge, token counts, prompt-cache hit rate, per-session and
  daily cost, burn rate, elapsed time, connection state, and a fleet badge
  (toggleable in Settings). Token/cost segments are populated for Claude and
  Codex sessions.
• Model picker with Ollama routing. The spawn form can select Anthropic /
  OpenAI / Ollama Cloud / Ollama Local models; the Hub injects the matching
  ANTHROPIC_* / OPENAI_* environment variables per session, with no shell
  setup required.
• Voice input. Prompts can be dictated through Browser speech recognition or
  a local Whisper server, including a Windows x64 managed whisper.cpp installer
  and an iPhone-Safari → Hub → PC Whisper relay, with near-silence and
  hallucination-phrase filtering on the Hub side.
• Mobile / smartphone access. The Hub UI is responsive (single-column layout
  on narrow viewports, touch-sized controls, and a mobile key panel for
  Esc/Ctrl/arrows) and documents reaching it from a phone over an SSH local
  forward; no public exposure is required or supported.
• Outbound approval notifications (ntfy / webhook). In addition to Web Push,
  the Hub can POST approval notifications to an ntfy topic or a generic webhook,
  which still works when no browser tunnel is connected. The Hub token is never
  included in the payload.
• Token-less loopback access. Requests from a 127.0.0.1 origin can reach
  the Hub UI without the URL token, configurable via an allowed-hosts setting;
  remote and tunnelled access still require the token.
• uninstall subcommand. many-ai-cli uninstall [--purge] removes settings
  and logs, and with --purge also removes the binary itself.
• Unified launcher with connection profiles (cross-platform).
  many-ai-cli-launcher handles WSL, SSH serve, and SSH tunnel profiles
  from one profile file and can reconnect to resident remote Hubs without
  restarting the remote session. The launcher binary now ships for Windows,
  Linux, and macOS (in every release archive and in the deb/rpm/Homebrew
  packages): SSH serve/tunnel profiles work on all three, while wsl
  profiles remain Windows-only and report a clear error elsewhere.
• Workbench tab and session history. The Hub now keeps a SQLite-backed
  session store and exposes stored sessions, timeline events, summaries,
  redacted exports, prompt templates, task/policy notes, diagnostics, usage
  summaries, stale-session views, file context helpers, and worktree helpers.
• PWA and opt-in Web Push notifications. The web UI ships a manifest,
  icons, service worker, push subscription settings, local VAPID key storage,
  and approval notifications that omit the Hub URL token.
• Remote server / Docker deployment assets. GHCR image publication, loopback-only
  per-user compose samples, resource limits, health checks, and an idempotent
  aac-update.sh workflow support resident remote Hub operation.
• Files and Git operations. The Files tab can create folders, save text
  files with base-mtime conflict detection, and delete empty folders. The Git
  view can fetch refs, run git pull --ff-only, and run a guarded plain
  git push (force/tag/delete disabled, non-interactive, confirmation dialog
  showing branch name and ahead count) from guarded Hub endpoints.
• Additional provider coverage. GitHub Copilot CLI and Cursor Agent CLI
  approval patterns, slash-command resources, usage links, and instruction
  injection paths are documented and wired into the Hub UI.

Changed

• Renamed the project from any-ai-cli to many-ai-cli (binary, config
  directory ~/.many-ai-cli/, MANY_AI_CLI* environment variables, approval
  markers [MANY-AI-CLI], Hub banner, and all public docs/UI strings). The npm
  package name any-ai-cli was too similar to an existing package, so the
  project is published as many-ai-cli.
• README and release-operation docs now distinguish Mark-of-the-Web,
  SmartScreen, Smart App Control, organization policy, and checksum/signature
  verification for unsigned Windows builds, prefer package-manager installation
  when available, and keep the GitHub Releases zip as the manual fallback.
• The public README files now describe v0.3.0 features, platform validation,
  Docker deployment, PWA/Web Push behavior, and Go 1.25 build requirements.
• The Release workflow now lets GoReleaser's before-hooks perform the frontend
  install/build once, avoiding a duplicate npm ci + npm run build step.
• Windows resource metadata is bumped to 0.3.0 so executable properties and
  manifests match the release tag.
• Browser-side third-party notices now match the xterm 6.0.0 package versions
  used by the current frontend type/dependency set.

Fixed

• Switching between sessions stays responsive after long-running connections:
  per-session pending output is now capped so a backlog no longer makes the
  first paint after a tab switch sluggish.
• Terminal output is no longer duplicated when the PTY size and the xterm.js
  render size briefly disagree during a window resize.
• Numerous terminal, approval, and voice UI refinements across xterm rendering,
  action-bar positioning/clearing, multi-pane interactions, Codex/Claude prompt
  detection, and voice diagnostics.
• GoReleaser no longer creates a Windows arm64 archive containing only
  the launcher binary and no matching main binary; Windows release
  archives remain x64-only until both binaries support arm64 together.
• SSH tunnel profile URLs now URL-encode Hub tokens for /api/info,
  /api/net-hint, and the browser URL, so manually configured tokens cannot
  break query parsing.
• The Hub status/probe URLs now also URL-encode tokens before local HTTP
  requests.
• Terminal output filtering now handles synchronized update control sequences
  without batching repaint output until the next user input.
• Public Docker samples no longer contain a personal user/service name.

Removed

• any-ai-cli-wsl.exe (standalone WSL launcher). The unified Windows
  launcher any-ai-cli-launcher.exe fully replaces it: a wsl profile
  provides the same behavior (bash -ilc startup, ANY_AI_CLI_WSL_LAUNCHER=1,
  automatic port selection, browser open, and WSL-side orphan cleanup). The
  Windows release zip now ships any-ai-cli.exe and any-ai-cli-launcher.exe.