Skip to main content

Changelog

All notable changes to AdaL CLI will be documented in this file.

[1.7.4] - 2026-09-03​

  • Smarter skill discovery — the agent searches the @skills hub (https://atskills.one) on its own when a task calls for specialist knowledge. Matched skills show a card you can view, save, or install inline.
  • @Skills Hub launched — discover, install, and manage reusable agent skills from the CLI or the full marketplace at https://atskills.one. Copy a skill path from the site, paste it in the CLI. Install less, automate more.
  • Native Smart Worktree Support - ask the agent to create a worktree and open a PR, the agent would create a worktree by itself without messing up the local filesystem
  • New models: Claude Fable 5.1, Gemini 3.8 Flash, Muse Spark 1.3
  • /model now shows available image and video generation models so you can see exactly what the agent uses for creative work.
  • Auto-update no longer interrupts running sessions — old versions are cleaned up only after every process using them has exited.

[1.7.3] - 2026-08-28​

  • @Skills Hub launched — discover, install, and manage reusable agent skills from the CLI or the full marketplace at https://atskills.one. Copy a skill path from the site, paste it in the CLI. Install less, automate more.
  • Optimized first launch startup speed - don't need to wait for too long for starting AdaL
  • New models: DeepSeek V4 Flash Vision (recommended), GLM-5.3 Flash, Qwen 3.8 Flash
  • More reliable image generation and tool use — arguments are validated before execution, preventing silent failures.
  • /resume now shows when each session was last active and how many messages it contains.

[1.7.2] - 2026-08-20​

  • @Skills Hub launched — discover, install, and manage reusable agent skills from the CLI or the full marketplace at https://atskills.one. Copy a skill path from the site, paste it in the CLI. Install less, automate more.
  • New models: GLM-5.3 (recommended flagship) and Qwen 3.8 Max (50% off through August)
  • Smarter cache reminders — the compaction nudge only fires when your context is actually filling up, not on every turn
  • Softer Default Dark theme with readable text selection across all themes
  • Credit usage warnings alert you before hitting your weekly limit
  • Session resume correctly tracks context size — /compact triggers reliably after resuming large sessions
  • Skills search returns relevant results for multi-word queries

[1.7.1] - 2026-08-17​

  • @Skills Hub launched — discover, install, and manage reusable agent skills from the CLI or the full marketplace at https://atskills.one. Copy a skill path from the site, paste it in the CLI. Install less, automate more.
  • Faster optimized booting time of the app
  • Enhanced onboarding flow of the website and app

[1.7.0] - 2026-08-13​

  • @Skills Hub is live — explore the full marketplace at https://atskills.one, discover, install, publish, and manage reusable agent skills right from the CLI or marketplace. Just copy paste the skills path from the website into the CLI. Install less. Automate more.
  • New model: Grok 4.6, Deepseek V4 Pro

[1.6.3] - 2026-08-08​

  • Fixed screen cutoff : All dialogs now scroll smoothly on any terminal size, with a height-adaptive layout
  • New Terminal Dark and Terminal Light themes — transparent background lets your terminal's own colors show through
  • Long lines in diffs, code blocks, and tool output no longer get cut off — proper character-level wrapping everywhere
  • Footer now shows your connection mode (BYOAK or ChatGPT Subscription) so you always know which key is active
  • Premium model warning appears after switching to high-cost models like Claude Opus or GPT-5.6 Sol
  • New model: Muse Spark 1.2 available in the model switcher

[1.5.7] - 2026-08-04​

  • /resume loads instantly even with large session histories, and adapts to any terminal size
  • Bring Your Own API Key now works for all providers — MiniMax, ZAI, DeepSeek, Kimi, XAI, Qwen, and Meta join the lineup
  • New model: Qwen 3.8 Max available in the model switcher
  • Right-edge rendering cleaned up — no more gray strip or uneven message padding
  • Up-arrow recall now shows what you actually typed, not expanded file contents
  • /web no longer 404s after recent internal build changes

[1.5.6] - 2026-08-01​

  • Each model now remembers its own reasoning effort level — switching models no longer resets your preference
  • Free-tier sessions self-heal after hitting a model restriction instead of staying stuck
  • Desktop applies updates automatically before launch — no more running stale builds

[1.5.5] - 2026-07-30​

  • Improved stability of UI
  • Default model upgraded to Claude Sonnet 5
  • Fixed issue where desktop first-time signup was not possible
  • /effort now takes effect on Gemini models; image generation migrated to Google's latest endpoints

[1.5.4] - 2026-07-29​

  • Optimized Memory Usage
  • Improved stability of UI
  • Backend auto-recovers mid-session crashes instead of leaving you stuck

[1.5.3] - 2026-07-24​

  • Added Claude Opus 5 — Anthropic's most capable model for complex agentic coding
  • Browser no longer steals focus when AdaL drives it; a control banner keeps you informed
  • Faster startup speed
  • /permissions works while the agent is streaming

[1.5.2] - 2026-07-23​

  • New UI : Made Easier and Simpler
  • Improved stability of UI
  • Qwen models now work correctly across all installation methods

[1.5.1] - 2026-07-21​

  • New UI : Made Easier and Simpler
  • Tool progress indicators recover correctly after interrupted responses
  • Fixed a rare freeze when searching very large codebases

[1.5.0] - 2026-07-20​

  • New UI : Made Easier and Simpler
  • Tool calls now fold into compact expandable sections — Click or use Ctrl+R to toggle
  • Reasoning effort level now displayed in the footer for supported models
  • Model picker simplified to a single Recommended section for faster selection
  • Kimi K3 (1M context) now available via /model
  • GPT-5.6 Terra is the new default model

[1.4.6] - 2026-07-12​

  • OpenAI GPT-5.6 family now available — Sol (frontier), Terra (balanced), and Luna (high-volume) via /model
  • Muse Spark reasoning steps now display correctly in conversations
  • Web and desktop apps rebuilt for faster startup and more reliable sessions

[1.4.5] - 2026-07-09​

  • Bug fix: Engineer mode workers now receive tasks instantly and reliably
  • xAI Grok 4.5 and Meta Muse Spark 1.1 now available via /model
  • Muse Spark 1.1 reasoning steps now stream live in the conversation

[1.4.4] - 2026-07-09​

  • xAI Grok 4.5 and Meta Muse Spark 1.1 now available via /model

[1.4.3] - 2026-07-08​

  • More reliable agent responses on longer conversations

[1.4.2] - 2026-07-07​

  • Engineer mode now launches seamlessly into tmux with no manual setup required
  • More reliable engineer mode for installations

[1.4.1] - 2026-07-04​

  • Smarter engineer mode — delegate tasks to supervised parallel workers with --mode engineer. The agent now adapts its process depth to match task complexity, skipping unnecessary steps on simple tasks and adding rigor on complex ones
  • More reliable conversations when using ZAI models with image-heavy history

[1.4.0] - 2026-07-04​

  • Engineer mode — delegate tasks to supervised parallel workers with --mode engineer
  • Multi-model consulting — the agent can now query multiple AI models in parallel and synthesize their answers for better results
  • More reliable installation and first-run experience across all platforms
  • Smoother onboarding flow

[1.3.0] - 2026-07-01​

[1.2.4] - 2026-06-12​

  • Claude Fable 5 now available via /model — Anthropic's new flagship with extended thinking
  • Desktop Home is workspace-first — pick a folder and press Enter to jump straight into a session
  • Improved light theme readability for code blocks and diffs across CLI and Web
  • File path actions in rendered markdown — click to open the source file or jump to Review
  • Side-by-side diff viewer now shows +/− markers for clearer change tracking
  • Desktop Review tab resumes reliably after reconnect — no more stale fallback states
  • File operations (delete, rename, move) in the Desktop file tree work correctly for all path formats
  • More accurate file-count badge on the current PR section

[1.2.3] - 2026-06-03​

  • Claude Opus 4.8 now available via /model — 1M context window with extended thinking
  • New --thinking-effort flag in headless mode (adal -q) — control reasoning depth per invocation from scripts and CI
  • Dotfiles (.env, .adal, .vscode) now visible in the Desktop and Web file tree sidebar
  • /agent is now the single command for all agent configuration — cleaner, one entry point
  • Unified /compact behavior across CLI, Desktop, and Web
  • Desktop sessions stay alive through screen sleep, lid close, and window minimize
  • Smarter browser automation — recovers automatically from stuck page sessions instead of giving up

[1.2.2] - 2026-05-28​

  • New /agent command — switch between Coding, Deep Research, and Browser Use modes from a single dialog
  • Shift+Tab now cycles through all permission modes: Strict → Accept Edits → YOLO
  • Tab key now toggles Plan mode only when in Coding mode
  • More reliable agent responses — eliminated a class of premature turn stops
  • Cleaner command cancellation — child processes no longer linger after Ctrl+C
  • Desktop: Full image support — paste, drag-and-drop, or send image-only queries
  • More reliable /resume — sessions restore cleanly without stale state errors
  • Removed deprecated Gemini 3 Pro from the model list

[1.2.1] - 2026-05-23​

  • Gemini 3.5 Flash model now available via /model
  • CLI: Context breakdown details now visible to all users in /stats session statistics
  • Desktop: Marked desktop as preview in header
  • Desktop: Seamless /ide handoff — Desktop app stays connected without backend interruption
  • Desktop: Clickable file paths in chat messages — click any inline path to open it in the file viewer
  • Desktop: Improved file viewer with better media previews and smart link navigation
  • Desktop: Smoother Desktop updates — progress shown in-app instead of native modal dialogs
  • Desktop: More reliable sessions after auto-updates (no more stale certificate errors)

[1.2.0] - 2026-05-20​

  • (Preview) Desktop App now available - download app and open with icon or /ide command
  • (Preview) Self-Review Agent available in Desktop App - open your PR and the agent auto-clusters and explains huge reviews for you
  • Smarter multi-turn conversations with extended thinking models
  • More reliable command execution with clearer error reporting when shell commands fail
  • Images you paste or upload persist correctly when resuming or switching sessions

[1.1.4] - 2026-05-15​

  • (Preview) /ide command now available - open your session in the browser with one keystroke
  • (Preview) self-review agent feature now available in /ide - open your PR and the agent auto-clusters and explains huge reviews for you
  • More reliable auto-updates on Windows
  • Improved instruction following and tool accuracy across all models
  • Improved cross-platform reliability on Windows with non-UTF-8 locales

[1.1.3] - 2026-05-09​

  • Unified tool expand/collapse toggle for cleaner message display
  • Faster first-query response times
  • Improved token tracking accuracy for all Chat Completions providers
  • Subagents run more reliably across providers

[1.1.1] - 2026-05-05​

  • Improved subagent performance and token efficiency
  • Fixed Anthropic multiple image input inconsistencies
  • Added support for reading SVG files
  • Added a native installer for Windows users
  • Added Browser Use mode; press Tab to toggle among Plan, Deep Research, and Browser Use modes

[1.0.9] - 2026-05-01​

  • Improved the UX of image gen with GPT and Gemini models
  • Fixed the windows crash bugs reported and improved product reliability

[1.0.8] - 2026-04-25​

  • Fixed image generation error and improved product reliability

[1.0.7] - 2026-04-25​

  • Added GPT-5.5 model support
  • First week 50%-off promotion for new models include GPT, DeepSeek, KIMI and Claude

[1.0.6] - 2026-04-24​

  • Fixed known issues in new model clients(DeepSeek & KIMI)

[1.0.5] - 2026-04-24​

  • Added GPT Image 2 support for image generation
  • Added DeepSeek V4 Flash&Pro models
  • Added KIMI 2.6 model
  • Minor UX polish and bug fixes

[1.0.4] - 2026-04-20​

  • Added support for Claude Opus 4.7 models
  • Minor UX polish and bug fixes

[1.0.3] - 2026-04-16​

  • Added Cron Agent support and interactive /cron management in the CLI
  • Fixed ChatGPT subscription streaming issues
  • Fixed Anthropic model image input handling
  • Improved error message handling

[1.0.2] - 2026-04-03​

  • Improved error messages display
  • Improved image generation requests stability

[1.0.1] - 2026-03-31​

  • Optimized the running memory of the AdaL session process
  • Improved Claude model requests stability
  • Added worktree info in exit UX if applicable

[1.0.0] - 2026-03-27​

  • Introduced Deep Research, a dedicated mode for investigating complex tasks and delivering structured reports (Tab to toggle)
  • Added Linux ARM64 support
  • Fixed Linux lagging issue
  • Enabled long-running agent execution when YOLO mode is toggled via /permissions
  • Improved streaming error handling and auto-retry
  • Added terminal notifications when AdaL needs user attention
  • Improved MCP user experience with smoother authentication and stable connections
  • Added a --resume flag for resuming previous sessions by ID
  • Resolved an issue affecting skill installation
  • Fixed slash-prefixed messages can't be queued mid-turn
  • Fetch tool display enrichment
  • Fixed an issue on WSL where pressing i for more info could fail silently
  • Improved chat history display with progressive loading
  • Fixed resumed-session display issues for subagents
  • Simplified compression report display
  • AdaL Web: Improved feature parity with AdaL CLI
  • AdaL Web: a redesigned UI
  • AdaL Web: added file view support
  • AdaL Web: Fixed browser not auto-opening on WSL

[0.9.9] - 2026-03-18​

  • Added MiniMax M2.7 and M2.7 Highspeed models
  • Added YOLO mode to auto-approve all tools and bash commands (/permissions)

[0.9.8] - 2026-03-17​

  • Improved Claude streaming reliability with fewer timeout errors
  • Improved error messages for rate limits for clarity

[0.9.7] - 2026-03-16​

  • Added GLM-5-Turbo from Z.ai
  • Reduced TimeoutError for Claude Opus models to allow long responses to pass

[0.9.6] - 2026-03-14​

  • Fixed MCP servers incorrectly showing as "not ready" after successful authentication

[0.9.5] - 2026-03-11​

  • Improved reliability for GLM requests

[0.9.4] - 2026-03-11​

  • Improved image generation reliability
  • Fixed incorrect mapping of queued @file references to images
  • Corrected hidden-line counts in diff view
  • Resolved a plugin install issue affecting some marketplace sources
  • Consolidated adal --help information
  • Renamed adal workspace to adal worktree and simplified its usage
  • Added --model support for headless mode to specify models
  • Added a clear "Text-only" capability indicator in model selection for non-vision models
  • Minor UI improvements for assistant messages

[0.9.3] - 2026-03-07​

  • Terminal tab title now shows AdaL
  • Bug fix in edit file path resolution

[0.9.2] - 2026-03-06​

  • Improved subagent performance
  • Added automatic subagent folding after completion
  • Further optimized prompt caching
  • Added @ autocomplete directory suggestions
  • Refined /model panel UI

[0.9.1] - 2026-03-05​

  • Added GPT-5.4 in ChatGPT subscription
  • Optimized cache hit rate
  • Optimized error self-recovery and subagent performance
  • AdaL Web: aligned stream/subagent behavior with AdaL CLI

[0.9.0] - 2026-03-05​

  • Added subagent support to improve AdaL's efficiency and performance
  • Optimized prompt caching to make AdaL 2x cost-effective
  • Added GPT-5.3 Codex and GPT-5.4 from OpenAI and set GPT-5.3 Codex as the current default model
  • Added GLM-5, GLM-4.7 FlashX and 3 more models from Zai
  • Upgraded image generation to Nano Banana 2 and enabled image editing
  • Memory optimization improvements for smoother long sessions
  • Corrected the effective input context window for all OpenAI models from 400k to 272k
  • Bug fixes in Ollama local model support
  • Refined /model panel metadata, pricing display, and selector UX
  • Removed /auth and merged auth details into /about
  • Keyboard shortcuts now respond to both lowercase and uppercase keys
  • Slash autocomplete completes commands without auto-running, allowing arguments to be added before execution
  • Fixed and optimized scrollbar behavior
  • Added context percentage display in the footer
  • Fixed inaccurate line counts in “click to expand”

[0.8.5] - 2026-02-28​

  • Fixed GPT models via ChatGPT subscription returning "Unsupported parameter" errors

[0.8.4] - 2026-02-27​

  • Fixed GPT models crashing during long sessions
  • Improved context window utilization to enable longer conversations before compaction
  • Deprecated older OpenAI models (GPT-4.1, o4-mini, standard GPT-5/5.1) and prioritized Codex variants
  • Added GPT-5.2 Codex as a recommended model

[0.8.3] - 2026-02-25​

  • Added image generation tool via Google Gemini's Nano Banana
  • Supported using ChatGPT subscription in adal to use OpenAI models
  • Added headless mode to run adal in non-interactive CLI execution (adal -q "query")
  • Supported select text to copy on all major platforms and terminals
  • Supported copy images from screenshots
  • Unified design on click to expand and collapse
  • Move status notification to display on the same row as the loading indicator for a more compact UI layout

[0.8.2] - 2026-02-21​

  • Faster MCP startup
  • Sync MCP tools when switching models

[0.8.1] - 2026-02-21​

  • Added read image tool to analyze images directly
  • Improved file editing with fewer retries and more reliable edits
  • Local model support via ollama (preview)

[0.8.0] - 2026-02-19​

  • Improved agentic tool use and communication across all models
  • New models - Google: Gemini 3.1 Pro
  • New models - MiniMax: M2.5 and M2.5 Highspeed
  • New models - Anthropic: Claude Sonnet 4.6
  • New models — OpenAI: GPT-5.2 Codex and 9 more
  • Redesigned /model dialog with provider-specific sections
  • Supported click to expand and collapse for lengthy display
  • Improved diff view with inline change highlighting
  • Overlay-style floating header on the top right corner
  • New theme selection with configurable terminal background
  • New user query UI design
  • Improved bash confirmation dialog layout
  • AdaL Web: real-time git branch updates and auto-refresh file explorer on branch change
  • AdaL Web: diff viewer in the sidebar for reviewing changes

[0.7.1] - 2026-02-12​

  • Improved @ search by skipping autocomplete when a space follows @
  • Fixed file editing issues on Windows
  • Reduced hallucination and improved error self-recovery
  • Bash confirmation dialog now shows agent description for better context
  • AdaL Web: smart scrolling with automatic scroll-to-bottom on query submission
  • AdaL Web: added AGENTS.md creation in the sidebar

[0.7.0] - 2026-02-10​

  • AdaL Web (Preview): a new browser-based interface for AdaL that can access core capabilities (adal --web)
  • Blazing-fast file search performance
  • Smarter auto-retry for model provider issues
  • Recommended models section in /model for quick access to top picks

[0.6.3] - 2026-02-05​

  • Added Claude Opus 4.6 (1M context window)
  • Always-on adaptive thinking that removes the thinking toggle
  • Loading indicator UI improvements
  • Model tags and display name cleanup

[0.6.2] - 2026-02-05​

  • Added Claude Opus 4.6 day 1 support (/model)
  • Improved error handling and retry experience
  • Smoother tool cancellation experience
  • Improved workspace management to enable parallel development
  • Optimized UI display for input area and toggle displays

[0.6.1] - 2026-02-02​

  • Simplified thinking mode from 3 modes to 2 modes (on/off)
  • Better scrolling experience with automatic scroll-to-bottom on query submission
  • Automatic retry logic for network errors
  • Better error handling and recovery
  • Minor bug fixes

[0.6.0] - 2026-01-31​

  • Parallel tool calls for ~50% faster execution and lower token cost
  • Lightning-fast memory compaction averaging 8s with more effective context retention
  • Better error handling with faster failure on context overflow
  • More accurate @ reference search navigation
  • Improved agent memory management
  • Improved session resume with consistent thinking content and message counts
  • Skills auto-update and improved UI
  • Supported using skills CLI to install skills to AdaL
  • Changed default model to Claude Opus 4.5
  • Clearer /init onboarding experience
  • Added 'i' keyboard shortcut in dialogs to open documentation
  • Supported Mermaid diagram rendering
  • Fixed table border misalignment and code block highlighting in answer rendering
  • Added current model name below the input box

[0.5.4] - 2026-01-28​

  • Minor bug fixes for compaction and session resume

[0.5.3] - 2026-01-27​

  • Faster @ reference search navigation
  • Workspace management to enable parallel development
  • Visual separator marker showing where compaction occurred when resuming sessions
  • Ctrl+R (thought toggle) and Ctrl+C (cancel/exit) now work when dialogs are open
  • Show changelog after auto update

[0.5.2] - 2026-01-24​

  • Thinking mode on by default
  • Better view when resuming sessions
  • Removed redundant aliases for a tidier command palette
  • Added spinner for bash tool execution display
  • Clearer plan mode separators in message history
  • Unified notification toasts for copy actions
  • Theme improvements
  • Other minor bug fixes

[0.5.1] - 2026-01-15​

  • Improved scrolling experience across all platforms
  • Smarter code display with automatic line wrapping in diffs and code blocks
  • Fixed known bugs in /model and /theme selections
  • Consolidated keyboard shortcuts and slash commands
  • Use Ctrl+C to cancel streaming anytime
  • Git-aware file operations that preserve history, better multi-line commit messages

[0.5.0] - 2026-01-13​

  • New UI: more stable, no flashing, no flickering, faster performance throughout the session
  • Improved input experience: cursor navigation, Shift+Enter for multi-line input, better query history navigation via up/down arrow
  • New logo and header design
  • Simplified /theme
  • Better markdown formatting and table rendering
  • Enhanced image understanding
  • Improved /compact
  • Enhanced /changelog with more history and link to documentation
  • Fixed auto-edit UI not updating after Shift+Tab or option selection during edit confirmations

[0.4.0] - 2026-01-07​

  • Supported more Linux distributions
  • Added Plan Mode (Ctrl+P) for planning-first workflows
  • Added google-gemini-3-flash-preview (/model)
  • Removed google-gemini-flash-2.5 and gpt-4o

[0.3.5] - 2026-01-05​

  • Faster query response
  • Supported more general web content for websearch, including images, places, and more
  • Fixed issues of fetching URL content
  • Fixed the "unknown slash command" issues
  • Improved bug reporting (/bug)
  • Enhanced file path handling

[0.3.4] - 2026-01-01​

  • Full support for skills (/skills): plugins/marketplace, personal and project skills
  • Improved branch display with cleaner format showing directory and branch
  • More robust auto-compact
  • Faster and more token efficient manual compact (/compact)
  • Added GPT-4o model support with robust output parsing
  • Enhanced error recovery
  • Other minor bug fixes and improvements

[0.3.3] - 2025-12-24​

  • More reliable file editing
  • Reduced UI flickering
  • Cleaner explanation and answer display

[0.3.2] - 2025-12-22​

  • Improved URL content fetch success rate

[0.3.1] - 2025-12-22​

  • Faster bash command display with real-time updates
  • Cleaner and more readable answer formatting
  • Improved URL content extraction and processing
  • Faster web search
  • Better file editing experience
  • Improved the UI stability
  • Enhanced /help with clearer guidance
  • Other minor improvements and bug fixes

[0.3.0] - 2025-12-15​

  • Reduced token usage by ~20%
  • Full history HTML view (/stats): export and view complete conversation history in HTML format
  • Unified input behavior: consistent copy/paste experience across texts, files, and images
  • Improved session resume (/resume): better state preservation and recovery
  • Faster response time: optimized agent execution
  • Integrated Claude Opus 4.5 into model pool (/model)
  • Other minor bug fixes

[0.2.3] - 2025-12-04​

  • Fixed known streaming bugs
  • Improved user experience