Terminal

• Split-panel layout — 70% main panel + 30% scrollable thumbnail bar showing all terminals
• Multiple terminals per workspace — Powered by xterm.js with full Unicode/CJK support
• Agent presets — Pre-configured terminal roles: Claude Code, Claude Code (worktree), Codex Agent, or plain terminal
• Git worktree isolation — Spawn Claude agents in an isolated worktree to prevent destructive changes to your main working tree
• Tab navigation — Switch between Terminal, Files, and Git views per workspace
• File browser — Search, navigate, and preview files with syntax highlighting (highlight.js)
• Git integration — Commit log, diff viewer, branch display, untracked file list, GitHub link detection
• GitHub panel — Browse PRs and issues directly from the Git tab
• Snippet manager — Save, organize, search, and paste code snippets (SQLite-backed with categories and favorites)
• Markdown preview — Preview .md files in a dedicated right sidebar panel with live file watching and right-click context menu
• Worker panel (Procfile) — Run multiple processes from a Procfile in a single tab with combined log view and per-process start/stop/restart controls, inspired by Overmind
• Per-terminal prompt history — Access previous commands per terminal session

Claude Code Agent

• Built-in Claude Code via SDK — Runs the agent directly inside the app; no separate terminal needed
• Message streaming with extended thinking blocks (collapsible)
• Permission modes — Multiple levels of tool execution control:
  • Default — Approve each tool call individually
  • Accept Edits — Auto-approve file edits, prompt for other tools
  • Plan mode — Agent proposes a plan file first; approve the plan to auto-execute
  • Bypass — Full auto-approval (use with caution)
• Subagent tracking — See spawned subagent tasks with progress indicators, elapsed time, and stall detection
• Session resume — Persist conversations and resume them across app restarts
• Session fork — Branch off from any point in a conversation with pending prompt auto-send
• Rest/Wake sessions — Pause and resume agent sessions from the context menu to save resources
• Effort level — Configure default effort level (high/medium/low) for new agent sessions
• Auto-compact — Automatically compact context when token count exceeds a configurable threshold

Statusline

A configurable status bar at the bottom of the agent panel, with 15 items across three zones (left / center / right):

| Item | Description | |---|---| | Session ID | First 8 chars of SDK session ID (click to resume a past session) | | Git Branch | Current git branch name | | Tokens | Total input + output token count (click for context breakdown) | | Turns | Number of conversation turns | | Duration | Session duration | | Context % | Percentage of context window used (color-coded) | | Cost | Total session cost in USD | | Workspace | Current workspace name | | 5h Usage / Reset | 5-hour API rate limit usage and reset countdown | | 7d Usage / Reset | 7-day API rate limit usage and reset countdown | | Max Output | Maximum output tokens for current model | | Cache Eff. | Cache read efficiency percentage (click for cache history with per-turn cost breakdown) | | Prompts | Link to view prompt history |

Items can be reordered, colored, and toggled on/off via a drag-and-drop template editor in Settings.

Cache Cost Awareness

• Cache history — Per-turn cache read/write breakdown with cost calculations per model (Opus, Sonnet, Haiku)
• Cache TTL countdown — Optional floating badge in the top-right corner showing remaining time for 5-minute and 1-hour cache TTLs; updates every 30 seconds, only appears after 1 minute of idle
• Cache expiry warning — Pre-send confirmation dialog when >150k cached tokens have expired (>1 hour), preventing accidental full-price reprocessing

Account & Usage

• Multi-account switching — /switch to manage and switch between multiple Claude accounts
• Usage monitoring — Track API rate limits (5-hour and 7-day windows) via Anthropic OAuth or Chrome session key
• Context usage panel — Visualize token usage breakdown by category (code, conversation, tools, memory, MCP, etc.)

UI & Interaction

• Image attachment — Drag-drop or use the attach button (up to 5 images per message)
• Clickable URLs — Markdown links and bare URLs open in the default browser
• Clickable file paths — Click any file path in agent output to preview it with syntax highlighting and search (Ctrl+F)
• Ctrl+P file picker — Fuzzy-search project files and attach them to the conversation context
• Skills & Agents panels — Browse available slash commands and agent configurations in the right sidebar
• Markdown preview search — In-pane Ctrl+F search inside markdown previews and file previews
• Long MCP tool output collapse — Auto-collapse oversized MCP tool results with a one-click expand
• Notifications — Dock badge, sound, and system notifications on agent completion (configurable)
• Update notifications — Automatic check for new releases on GitHub

Codex Agent

Optional alternate agent backend powered by @openai/codex-sdk. Pick Codex Agent (or Codex Agent (worktree)) from the agent preset list when creating a terminal.

• GPT-5.5 / 5.4 / 5.3-codex / o4-mini / o3 / GPT-4.1 — Switch models inline; ChatGPT login or OpenAI API key
• Sandbox modes — read-only, workspace-write, or danger-full-access
• Approval policies — untrusted, on-request, or never (auto)
• Worktree preset — Spawn Codex in an isolated git worktree, same as the Claude worktree flow
• Session resume — JSONL transcripts under ~/.codex/sessions/ are auto-indexed; resume hits a local cache so re-opening a thread is fast even with months of history
• Inline plan approval — Codex plan items render as approvable blocks in the panel

OpenAI Direct (debug)

Hidden behind BAT_DEBUG=1. Talks directly to the OpenAI Responses API via @ai-sdk/openai, without the Codex CLI. Useful for testing tool flows against a raw model. Not recommended for daily use — Codex Agent is the supported path.

Internationalization (i18n)

• English, Traditional Chinese (繁體中文), Simplified Chinese (简体中文)

Keyboard Shortcuts

| Shortcut | Action | |---|---| | Ctrl+`` / `Cmd+ `` | Toggle between Agent terminal and first regular terminal | |Ctrl+←/→/Cmd+←/→| Cycle workspace tabs (Terminal / Files / Git) | |Ctrl+↑/↓/Cmd+↑/↓| Switch to previous / next workspace | |Ctrl+P/Cmd+P| File picker (search & attach files to agent context) | |Ctrl+N/Cmd+N| Open new window | |Shift+Tab| Switch between Terminal and Agent mode | |Enter| Send message | |Shift+Enter| Insert newline (multiline input) | |Escape| Stop streaming / close modal | |Ctrl+Shift+C| Copy selected text | |Ctrl+Shift+V| Paste from clipboard | |Right-click` | Copy (if text selected) or Paste |

Slash Commands

| Command | Description | |---|---| | /resume | Resume a previous Claude session from history | | /model | Switch between available Claude models | | /new / /clear | Reset session (clear conversation, fresh start) | | /abort | Stop the current agent session immediately | | /snippet | Show snippets to Claude for management | | /switch | Switch between Claude accounts | | /login | Sign in to Claude (switch a