v0.3.2

Added

• Built-in API proxy + Payload chat view (β). The Hub now embeds a
  transparent HTTP proxy for https://api.anthropic.com/v1/messages,
  https://api.openai.com/v1/chat/completions, and
  https://api.openai.com/v1/responses. Wrapped CLIs (Claude Code / Codex CLI)
  are pointed at the local proxy via ANTHROPIC_BASE_URL / OPENAI_BASE_URL
  env injection, so every request/response is captured per session without TLS
  MITM or CA distribution. A new "Payload 表示 (β)" toggle on the chat pane
  shows a clean, structured turn list sourced from the proxy (independent of
  PTY scraping), and each turn has a [Raw] toggle that opens the raw JSON.
  Authorization headers are stripped before the payload reaches the in-memory
  sink — credentials are never persisted.
• Tailscale wizard: inline "Enable external exposure" button. The mobile
  connect wizard step ④ now ships a one-click button that flips the Tailscale
  serve toggle from inside the wizard instead of bouncing the user back to the
  toolbar.
• README (English / Japanese) now flags the smartphone section as a
  beta / draft preview shipped in v0.3.x.

Fixed

• Approval popup garbage. ANSI escape leftovers in the action-bar question
  panel (approval-popup-stripansi-garbage), the missing option 1 in
  wrapped-fallback layouts (approval-fallback-wrapped-opt1-missing), the free
  input misalignment in inline-right layouts (approval-free-input-inline-right),
  the Q-tab shrinkage when a batch preamble was long
  (batch-preamble-shrinks-qtabs), and the preamble's table/rule garbage are
  all addressed by moving the preamble inside the marker and stripping any
  rules/tables that slip through.
• Token statusbar / cost popover. The cost popover no longer ends up
  visually behind the statusbar (token-statusbar-cost-popover-under-statusbar).
• Live status palette popover. No longer clipped at the terminal edge
  (live-status-palette-popover-clipped).
• Spawn cwd. Typed paths are now existence-checked before the session is
  created (spawn-cwd-existence-unchecked-for-typed-path).
• Session card order. Eliminated the random-snapshot reorder that swapped
  card positions on refresh (session-card-order-random-snapshot).
• Card branch badge. Restored right padding on the branch badge
  (card-branch-badge-right-padding-gap).
• AI commit message marker. Stripped the stray bullet prefix that leaked
  into the generated commit message (commit-ai-marker-bullet-prefix).
• Project group grid button style. Aligned with the rest of the grid
  (project-group-grid-btn-style).
• Quick-command settings. Restored vertical breathing room
  (quick-cmd-settings-vertical-compress).
• Settings confirm. Confirmation now uses the toast pattern instead of
  closing the panel (settings-confirm-closes-panel-toast-unify).
• Workflow modal. Spinner now settles correctly
  (workflow-modal-spinner-not-settling); progress bar percentage is readable
  again (workflow-progress-bar-pct-unreadable).
• Mobile statusbar overflow. Long status text no longer pushes the
  statusbar off the viewport (mobile-statusbar-overflow).
• Mobile WebSocket auth. The mobile flow now falls back to the cookie when
  the URL token is unavailable (mobile-ws-token-cookie-fallback).

v0.3.1

Fixed

• The approval action bar no longer renders a tall empty column to the left of
  the question panel. In the batch / single-question / multi-select layouts the
  APPROVAL NEEDED label kept flex-basis: 100% (intended for the wrapped row
  layout), which in the flex-direction: column variants behaves as a
  100%-height basis and pushed the question tabs/panel into a second wrapped
  column. The label now uses flex-basis: auto and the column layouts set
  flex-wrap: nowrap.

Changed

• The Cursor Agent slash-command resource
  (resources/slash-commands/cursor-agent.md) is now populated from the
  official Cursor CLI documentation (about 22 commands such as /plan, /ask,
  /model, /auto-run, /sandbox, /compress) instead of the previous
  4-command placeholder.

Full Changelog: v0.3.0...v0.3.1

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.

v0.2.2

Changelog

v0.1.3

Changelog

• af59892 fix: Codex 承認検出とバージョン表示を改善
• 0d4f787 fix: 音声入力のマイク取得干渉を回避
• 4df1762 merge: develop を v0.1.3 として main に反映

v0.1.2

Changelog

v0.1.1

Changelog

• f293577 fix: enforce LF for THIRD_PARTY_NOTICES.md and vendored licenses
• 5b85c21 fix: normalize line endings in THIRD_PARTY_NOTICES generator