- @ Context Picker: Type
@in the input to browse and attach agents and rules files to your prompt - Marketplace Auto-Update: Installed marketplace repos are automatically updated in the background on startup; toggle in marketplace settings.
- Recommended Marketplaces: Projects can declare recommended marketplaces in
.augment/settings.jsonthat prompt installation on first open. These must be in the project level settings.json file - MCP OAuth Paste Mode: When authenticating MCP servers over SSH, you can now paste the auth code or redirect URL directly instead of relying on localhost redirect
- Effort Levels: Configure agent effort/reasoning levels for chat requests
- Enter Plan Mode Tool: The agent can now enter plan mode mid-conversation when appropriate
- Deprecated various flags for auth - ENV variables AUGMENT_API_TOKEN, AUGMENT_API_URL and flags
--augment-token-fileand--api-urlare no longer supported - Use
AUGMENT_SESSION_AUTHand--augment-session-jsoninstead
- Parallel Tool Execution: The CLI agent loop now executes independent tools in parallel for faster responses
- Plan Mode: Plans are now saved to
~/.augment/plans/and plan mode enforces strict read-only access - Notification Bell on Ask-User: Terminal bell now rings when the agent prompts for user input (when
notificationModeis set tobell) - Custom Command History: Custom slash commands are now saved to input history for easy recall with the up arrow
- Models List:
auggie models listnow shows full model information --helpOutput: Added--queue,--show-credits, and--mcpto the CLI help text--augment-session-jsonFlag: Now accepts both inline JSON and file paths (e.g.,--augment-session-json ~/.augment/session.json)- Conversation Retrieval: Improved relevance ranking and caching for the conversation retrieval tool
- Terminal Tool: Better detection of commands that block waiting for stdin input
- Jira/Confluence Errors: Validation errors from Jira and Confluence are now surfaced in chat instead of showing a generic failure message
- Fixed
/mcptoggle not actually enabling/disabling MCP servers - Fixed cursor position after the
/clearcommand - Fixed file write/delete silently failing in multi-folder workspaces
- Fixed queue display showing duplicate messages
- Fixed auggie command [name] syntax for executing custom commands
- Multi-workspace support: work with multiple folders simultaneously via
--add-workspaceflag or/add-workspaceslash command - Skills as slash commands: trigger skills deterministically via
/commands instead of relying on agent auto-detection - Multiple settings files: choose which settings file to save to when multiple exist, with 🔒 indicators for organization-managed settings
- MCP OAuth scopes: MCP server authentication now supports configuring OAuth scopes
- Ask-user tool: replaced custom text input with a "Chat about this" option for more natural conversational clarification
- Ask-user shortcuts: press number keys (1–9) to directly select options servers
- Session picker: added keyboard shortcuts for faster navigation
- Processing duration: now displays minutes and hours instead of only seconds
- Input history: navigating history no longer opens the slash command menu
- Pasting: pasting text no longer auto-switches to ask mode
- Upgrade notification: now tells you to restart the CLI after upgrading
- Session resume: resuming a session now preserves your originally selected model
- Tool permissions: default to denylist mode to prevent accidental lockout from all tools
- Fixed CLI crash on startup caused by logger initialization order
- Fixed stuck queued messages in cloud mode
- Fixed apply_patch writing corrupted files when the model produced malformed output
- Fixed slow CLI exit on macOS/Windows (improved from ~27s to ~200ms)
- Model Picker Badges: Model picker now shows cost tier indicators and server-controlled badges (e.g. "Free", "New")
- Ask Mode Enhancements: Slash commands (
/) and external editor (Ctrl+O) now work in Ask mode - Single-Click Login: Streamlined browser-based authentication flow
- /context:
/contextto see context window usage
- Sub-agents Credit Usage: Add support for subagents credit usage via non interactive mode for enterprise customers and /stats for all users
- Faster Tool Execution: Reduced tool execution delays in large workspaces by caching enumeration state and parallelizing rule file reads
- Incremental Session Saving: Agent progress is now saved after each LLM exchange, preventing loss of work if the process crashes mid-turn
- Message Queue Persistence: Queued messages are now saved to the session file, so they survive CLI restarts
- Smarter Input Modes: Pasting or recalling history entries with a prefix (
/,!,#) now automatically enters the corresponding input mode - Slash Command Ordering: Exact alias matches (e.g.
/q→queue) are now prioritized over fuzzy matches - Stash/Recall Hints: Added Ctrl+S (stash) and Ctrl+T (recall) to help screens for easier discovery
- Task Tool Performance: Parallelized task tool operations to prevent occasional hanging
- Chat History Truncation: Improved content-aware truncation measurement with a higher default limit, preventing premature history loss in long sessions
- Network Resilience: Improved error messages and circuit breakers for unstable network connections
- Ask mode improvements: Slash commands and Ctrl+O (external editor) now work in ask mode
- Ctrl+R history search: Persistent reverse history search across sessions
- Selective session deletion: Delete individual sessions from the session picker
- Message queue editing: Up arrow key opens the queue popover, with a hint shown in the message placeholder
- Custom slash commands in --queue: Queued messages now resolve custom slash commands (e.g.,
--queue "/my-command") - --queue support in non-interactive mode: Queue multiple prompts in
--printmode for chained workflows
- Indexing denylist: Permanently decline indexing for specific workspaces with a "Never index this workspace" option
- Incremental session saving: Agent progress is now saved after each LLM exchange, preventing work loss on crashes
- Task management: Task system now auto-recovers if the task list is missing, and task names are shown in update displays
- Sub-agents now have access to MCP tools from their parent agent
- Session auto-rename now triggers correctly for messages with tool calls
- Config wizard: All options now visible without scrolling
- Session list ordering: Newest sessions appear first
- User message display: Improved formatting with full row highlight and brighter text
- Improved summarization with incremental updates and backend-driven token counting
- Remote history sync: Better conflict resolution for synced sessions
- Memory improvement during git indexing
- MCP OAuth: Better handling of non-conformant server responses for dynamic client registration
- TUI rendering flickering improvements
- Localhost OAuth login: local sessions now authenticate via browser-based OAuth flow instead of JSON paste
- Session naming: name of sessions via
/renamecommand is now displayed to the user - Model picker search: Option+M hotkey opens the model picker, which now supports search/filter
- Prompt stashing: press Ctrl+S while typing to stash your prompt and recall it later
/statscommand: view session billing and usage details- MCP server toggling: enable/disable individual MCP servers from the MCP popover
- MCP log streaming: MCP server logs are now visible in the TUI for easier debugging
- MCP token variable:
${augmentToken}variable expansion available for MCP server headers .agentsdirectory: added support for.agentsdirectory for skill and agent discovery- History summarization indicator: visual indicator shown when conversation history is being summarized
- Hierarchical rules indicator: visual indicator showing active AGENTS.md rules
- Auth flags: added
--augment-session-jsonflag andAUGMENT_SESSION_AUTHenv var as recommended auth methods (old flags deprecated but still work) - MCP compatibility: improved compatibility with non-standard MCP server JSON schemas (e.g., mcp-server-terminal)
- View tool display: correctly shows "read directory" with entry count instead of "read file" with "0 lines"
- Image attachment indicator moved closer to the input textbox
- Removed distracting "Your queue is done!" popup
- Removed misleading "To see what's new" message after upgrade
- Fixed Ctrl+C not exiting the CLI on macOS (no longer requires
kill -9) - Fixed crash on exit on Windows (UV_HANDLE_CLOSING assertion)
- Fixed crash when pasting text or using Ctrl+P prompt enhancement
- Fixed
/logoutrequiring two attempts to fully log out - Fixed built-in subagents (explore, plan) disappearing after config changes
- Fixed sub-agents hanging indefinitely during codebase retrieval
- Fixed interleaved/garbled streaming output when sending messages rapidly
- Fixed Option+Backspace word deletion in kitty protocol terminals
- Fixed Ctrl+W word deletion not treating newlines as word boundaries
- Fixed verbose mode truncating the first line of bash command output
- Fixed
--quietflag not suppressing MCP server initialization messages - Fixed MCP server OAuth authentication not responding to Enter key
- Fixed session resume failing after workspace switch
- Fixed
/newcommand in cloud agent mode not creating a new session - Fixed message queue stalling until a new message was sent
- Fixed spurious warnings when settings.json is empty
- Fixed prompt hint color changing when text wraps to a new line
- Fixed custom command parameter hint not disappearing after typing a space
- Fixed text wrapping issues at narrow terminal widths
- Fixed
auggie tools removenot showing an error for non-existent tools - Fixed sub-agent failures showing "Done in 0s" instead of error details
- Fixed numpad keys not working correctly
- Improved error messages when resuming sessions
- Agent Skills Support: Added support for loading specialized domain knowledge from SKILL.md files following the agentskills.io specification
- Skills Viewer: Added
/skillscommand to display currently loaded skills and approximate token usage - Prompt Enhancement: Added
--enhance-promptflag for non-interactive mode to improve prompts before sending to agent
- Session State Preservation: Workspace settings (guidelines, rules, memories) are now preserved when using
/newcommand or--continueflag - Hook Message Display: Hook messages now appear inline after each tool result instead of being batched at the bottom for better context
- Session Picker Ordering: Session picker now displays most recent sessions at the top of the list
- Bash Mode Display: Bash mode output now appears muted while running and displays in full when complete
- Exit Shortcuts: Improved exit shortcuts (Ctrl+C, Ctrl+D, Escape) to work consistently from any popover state
- Custom Command Model Override: Custom command model overrides now only apply to the next response instead of all follow-up responses
- Chat History Display: Chat history now only shows user message entries for actual user input, not system-generated content
- Session Resumption: Fixed tool results (ViewTool, EditTool) not rendering when resuming sessions via
--resumeor/sessionscommand - Keyboard Navigation: Fixed keyboard shortcuts not responding while in mention mode
- Popover Input Handling: Fixed keyboard input handling in popover states
- Input Focus Characters: Fixed issue with focus characters appearing in input
- Queue Mode Shortcuts: Updated queue mode keyboard shortcuts - X now deletes items, D moves items down
- MCP dynamic workspace discovery: MCP mode now supports on-the-fly workspace indexing with --mcp-auto-workspace flag, allowing codebase-retrieval to search different directories without restarting
- Session management: Most recent sessions now appear at the top of the session picker list
- Chat history: Chat history display fixed on --continue
- Custom commands: Custom command model overrides now only apply to the next response instead of all follow-up responses
- Banner: Updated banner with left-to-right gradient and all caps text
- Settings management: Settings updates now preserve comments and invalid/unknown fields in settings.json
- Session persistence: Workspace settings (guidelines, rules, memories) are now preserved when using /new command or --continue flag
- Fixed keyboard shortcuts not responding while in mention mode
- Fixed issue with focus characters appearing in input
- Fixed rules not being applied when starting a new session with /new or --continue
- Fixed MCP server orphan processes when parent process terminates
- Onboarding: Updated prompt enhancer onboarding text to clarify credit usage
- Terminal title updates: Terminal window title now automatically updates based on conversation context
- Process lifecycle management: Background processes launched via shell tools and launch-process are now tracked in /bash viewer and automatically cleaned up on exit
- Login improvements: Login flow now preserves existing session if authentication fails, and browser opening has a timeout to prevent hanging on remote machines
- Terminal-aware session resumption: Using -c flag now prefers the most recent session from the current terminal
- Model selection: CLI now automatically selects the default model when no model is configured
- Print mode enhancements: Print mode now outputs request ID for easier debugging and support
- Session display consistency: Session display format is now consistent between /sessions list and session picker
- Subagent visibility: Subagent output now includes thinking summary for better visibility into agent reasoning
- Configuration options: Added auto-update configuration option to /config menu
- Rules interface: Improved rules popover interface with better navigation and display
- /about command: View user information and debugging details with a new slash command
- Keyboard navigation: Improved navigation with consistent arrow key and j/k shortcuts across all TUI components
- Task list UI: Fix uncaught exception that resulted in the agent being unable to add to the tasklist
- MCP Logging: Added --log-file option for MCP mode to enable error logging (default: /tmp/augment-log.txt)
- MCP Performance: Improved MCP mode startup performance with asynchronous workspace initialization
- Thinking Display: Improved visual styling for thinking entries with better readability and less intrusive appearance
- Bash Mode: Bash mode is now available, trigger by pressing !
- Input History: Fixed input history navigation state after submitting commands and queued messages now properly added to history
- Tool Interruption: Fixed crash when interrupting tool execution with escape key
- Parallel Interrupts: Improved interrupt handling to properly cancel all running parallel sub-agents
- Exit Message: Fixed typo in exit tip message
- Session Switching: Added
/sessionscommand to switch between sessions without restarting CLI - Shell Configuration: Added
/configcommand to configure default shell and startup script - Keyboard Shortcuts: Added Ctrl+/ for undo and Ctrl+Y for redo in normal edit mode
- Verbose Thinking: Added verbose output mode for thinking summaries to show full agent reasoning
- Terminal Focus Tracking: Added focus detection to hide cursor when terminal window loses focus
- Show session ID when closing sessions with command to resume by ID
- Support resuming sessions by ID prefix (unambiguous matches)
- Added
-fflag to filter session list to current workspace only - Reversed session list order to show newest sessions first
- Fixed text wrapping in tool result summaries
- Truncate web fetch results to 150 characters for better readability
- Increased max length for thinking summary titles to reduce truncation
- Truncate large file views to prevent excessive memory usage
- Fixed escape key behavior in ask mode (now properly exits to normal mode)
- Fixed race condition in terminal focus tracking that caused escape sequences to leak
- Fixed ACP login flow path identification
- Removed duplicate error messages in output
- Suppressed verbose npm output during auto-upgrade failures for cleaner error messages
- Fixed issue with extraneous git processes spawning after indexing
- Session Sharing: New
/sharecommand in TUI andaugment session shareCLI command to generate shareable links for chat sessions - Auto-Update Control: New
autoUpdatesetting in settings.json to control automatic updates
- ACP Mode: Now fully released (no longer experimental) with non-interactive chat mode and thinking summaries for better visibility
- TUI Performance: Improved rendering performance with Ink 6.5.0 incremental rendering
- Session Resumption: Chat history now displays when resuming sessions with
--continueor--resumeflags - Agent Capabilities: Enhanced apply_patch tool with more robust patch parsing for better file editing reliability
- Error Messages: Improved error messages when file editing operations fail
- Shell Configuration: Added automatic shell detection and startup script support with
--shelland--startup-scriptoptions - Windows Support: Fixed PowerShell execution on Windows for better cross-platform compatibility
- Terminal Authentication: Added terminal authentication support for ACP mode
- ACP Mode: Made
--acpflag publicly available for Agent Communication Protocol support - Queue Management: Enabled message queue when agent is busy
- Rules Management: Improved rules filtering and status display for better visibility of manual and dynamic rules
- Chat Summarization: Fixed chat history summarization
- Agent Client Protocol (ACP) Support: Added comprehensive support for the Agent Client Protocol with terminal authentication, model selection, and indexing control via new
--acpand--allow-indexingflags - Thinking Summaries: Display GPT-5 model thinking process summaries with collapsible sections in the TUI
- MCP Server Support: Extended support for HTTP and SSE-based Model Context Protocol servers in addition to stdio servers
- CLI is now GA: Removed beta label from the CLI banner
- Model Selection: Simplified model selection with short names (e.g., 'sonnet4.5' instead of full model IDs)
- File Mentions: Fixed file mentions to not include @ symbol in prompts
- Content Handling: Improved stability when handling undefined content to prevent crashes
- Parallel Tool Calls: Added support for models calling multiple tools simultaneously
- Agent Client Protocol (ACP): Added experimental support for external editor integration via
--acpflag, including file mentions and image support - User Rules: Added support for user-specific rules in
~/.augment/rulesdirectory for custom agent behavior - Tool Management: Added
--disable-toolflag and settings configuration to disable specific tools from the agent's toolset
- Vim Mode: Added 'e' keybind for moving to the end of a word, matching standard vim behavior
- Session Picker: Improved UI with dynamic column resizing for better readability
- Settings Validation: Enhanced error handling to gracefully handle invalid configuration fields
- Request ID: Added
/request-idcommand to display request IDs for debugging and support
- Commands: Added
/request-idcommand to display request IDs for debugging - UI: Improved session picker with dynamic column resizing
- UI: Added modified time to session list display
- UI: Fixed credit usage display to round down for accuracy
- Settings: Improved settings validation to handle invalid fields gracefully
- Errors: Added request IDs to API error messages for better debugging
- Stability: Fixed crash when using @ with large codebases (150,000+ files)
- MCP: Fixed MCP server configuration validation to prevent crashes
- Performance: Fixed file picker performance issue that caused UI lag
- Image support: Attach images to prompts using
--imageflag or the new/imagecommand with drag-and-drop and paste support - Enhanced clipboard: Fixed clipboard copying in SSH sessions and terminal multiplexers (tmux/screen) using OSC 52 protocol
- Session improvements: Session list now shows last modified time and request IDs for better debugging
- OAuth improvements: Display authentication URL when browser fails to open (useful for SSH connections)
- Simplified commands: Consolidated
/newand/clearcommands for starting fresh conversations
- Added
/copycommand to copy request ID or response text to clipboard - Added OAuth authentication support for MCP (Model Context Protocol) servers
- Added interactive session picker when using
--resumewithout specifying a session ID - Improved help command readability with better formatting and organization
- Added fuzzy search for slash commands, making them easier to discover
- Improved file picker performance with better fuzzy search algorithm for large codebases
- Enhanced tool permission system reliability with improved regex matching
- Fixed extra blank lines appearing at the beginning of agent responses
- MCP Support: Added
mcp add-jsoncommand for importing MCP servers via JSON configuration - MCP Reliability: Improved MCP server validation to continue loading valid servers even when some configurations are invalid
- Version Management: Enhanced upgrade system with semantic versioning support and better handling of prerelease versions
- TUI Navigation: Improved input mode switching with history-based navigation for more natural mode transitions
- File Picker: Fixed double @ symbol display issue in file picker mode
- Select Menus: Fixed arrow key handling in TUI select menus for smoother navigation
- Non-Interactive Mode: Agent now runs without stopping for user input when in non-interactive mode
- Tool Execution: Fixed regex command execution bug that was causing incorrect string formatting
- Tool Permissions: Improved consistency in tool permission settings format
- Feedback: Added GitHub repository link to the feedback command for easier issue reporting
- New
auggie rules listcommand: Display workspace rules and guidelines directly from the CLI - Enhanced
/rulescommand: View workspace rules with improved colored formatting in the TUI - Smarter TUI mode switching: Exit modes now return to the previous mode instead of always going back to Normal mode, enabling better workflows with FilePicker and Ask modes
- Session continuation tip: See a helpful reminder about using
auggie session continuewhen exiting the TUI - Better error messages: Reduced duplicate warnings and improved error tracking for cleaner output
- Fixed Ctrl+C handling: Properly interrupt the agent when Ctrl+C is pressed while the agent is running
- Fixed Option+Delete: Keyboard shortcut now correctly deletes words backward instead of forward
- Fixed issue where rules were not being attached to the agent request
- Ask Mode: a streamlined prompt-first interaction mode with improved transcript rendering.
- New session commands:
auggie session listandauggie session resumeto manage and continue sessions. - MCP quality-of-life: simpler
mcp addsyntax, automatic migration of legacy settings, and a/statuscheck. - Add Ask Mode for CLI/TUI, with formatting and input history improvements.
- Add
auggie session listandauggie session resumecommands. - Add
--permissionflag to configure tool permissions at runtime. - Support compressed syntax for
mcp addand auto-migrate legacy MCP settings.json format. - Add
/statuscommand to check MCP and rules status. - Add
--max-turnsto cap agent iterations in print mode. - Add basic JSON output mode.
- Improve keyboard handling: Ctrl+C to clear input, Ctrl+D forward delete; fix Delete key acting as Backspace.
- Improve non-interactive error messages when auth is missing.
- Fix crash when denying indexing by properly initializing status messaging.
- Fix markdown list indentation in TUI output.
- Fixed issue where rules were dropped when combined content was too long.
- MCP settings now use a record-based schema; legacy formats auto-migrate on run.
- Manage Model Context Protocol (MCP) servers with
auggie mcp add|list|remove - Configure MCP servers in your settings file (
~/.augment/settings.json) or via a config file passed with--mcp-config - Skip waiting for indexing to complete before codebase retrieval executes in TUI mode
- If API requests are retried, the CLI shows a clear message so you know what's happening
- Interrupting an operation now cleans up any partial output to keep the screen tidy
- Custom slash command help text now shows the selected model; logging and parsing are consistent in both interactive and non‑interactive modes
- Session tracking is more reliable between the CLI and the API
- Authentication works correctly when you provide both an API token and an API URL
- On Windows, home‑directory detection across different drives has been fixed to avoid incorrect indexing
- Non-interactive slash command syntax: Run custom commands directly from the shell with
auggie /<command> ...(e.g.,auggie "/joke robot" -p) - Indexing is now enabled by default in print mode, with a safety guard that disables indexing when running from your home directory to avoid accidentally uploading your entire home folder
- Custom commands can now specify which AI model to use in their frontmatter configuration
- Slash commands can be run directly from the command line (e.g.,
auggie /help) - Improved paste functionality on Windows and support for Chinese character input
- Enhanced feedback submission with proper handling of bracketed paste
- Quiet mode (
--quiet) now automatically defaults to text output mode - Empty agent responses are no longer displayed in the CLI interface
- CLI continues gracefully when settings fail to load instead of crashing
- Invalid MCP configuration no longer causes crashes
- Added workspace size limits to prevent indexing excessively large directories
- Fixed typo: "enchance" → "enhance" in prompt enhancer
- Added ability to edit existing tool permission rules without recreating them (press 'e' to edit)
- Improved tool permissions UI with better visual hierarchy and clearer permission types display
- Simplified permissions management
- Added long help text that can be toggled with '?' in Normal mode
- Fixed Ctrl+C and double escape shortcuts to work properly in vim mode
- Automatically enter INSERT mode when typing
/vimcommand - Cleaned up vim mode help text for better clarity
- Added slash commands to the history manager for easier command recall
- Fixed command exit behavior - commands like
model list,tokens print, andsession deletenow properly exit after completion - Prevented indexing when running CLI from home directory to avoid performance issues
- Fixed git worktree detection
- Limited @ file cache size to prevent performance degradation when opened in large directories
- Display default model in brackets when no model is explicitly selected
- Updated onboarding messages and tips for better user experience
- Added
/editorcommand to compose prompts in your preferred external editor (VS Code, vim, nano, etc.) - Added
Ctrl+Okeyboard shortcut for quick access to external editor - Added current AI model name display in the status line footer
- Added
j/kkeyboard navigation support for all menus (Task List, Model Picker, GitHub workflow) - Improved popover heights for better visibility on smaller screens
- Fixed footer wrapping issues on narrow terminal widths
- Added
/verbosecommand to toggle between compact and detailed transcript views - Fixed verbose command toggle to properly apply changes in the current session
- Improved help text with dynamic command listing and better keyboard shortcut documentation
- Enhanced onboarding experience with clearer feature descriptions
- Implemented comprehensive theme-based system across markdown, tool formatters, and status indicators
- Reduced flicker in iTerm when opening/closing popovers
- Added smart git update events with file tracking for better performance
- Improved error handling in str-replace-editor tool to detect when old and new strings are the same
- Enhanced settings management with structured return types and better error handling
- Fixed double @ appearing in file picker mode
- Fixed indexing confirmation to wait before initializing runtime
- Fixed tool permission type display for "ask-user" permissions
- Corrected typos and improved help text clarity
- Fixed clipboard functionality with proper feedback and error handling
- Enhanced scrollable list components for better handling of long item lists
- Added support for custom slash commands (
/<command>) in interactive mode and--commandflag in non-interactive mode - Support for nested custom commands using colon separator (e.g.,
nested:command) - Custom commands can be defined in
.augment/commandsor~/.augment/commandsdirectories - Added tab completion for slash commands in interactive mode
- Custom commands from Claude Code are now automatically detected and imported
- Added detailed help section for
--commandflag withauggie --command --help - Added shorthand flags for common options:
-i(input),-if(input-file),-p(prompt),-q(query),-lm(list-models),-m(model),-cm(command) - Added user settings support via
~/.augment/settings.jsonfor persistent model preferences - Support for CLAUDE.md and AGENTS.md configuration files in addition to .augment/guidelines.md
- Settings are validated with schema and gracefully handle malformed JSON
- Added prompt enhancement via Ctrl+P in interactive mode to improve prompts using AI
- Improved navigation in input box with Option+Left/Right for word navigation
- Enhanced multi-line prompt navigation with up/down arrows
- Added repository and branch information badges to agent cards
- Improved select menu system with better modularity and pagination support
- Fixed regression in input history and cursor movement in interactive mode
- Fixed extraneous border display on successful side effects with green border