LoginDownload
/ changelog

Changelog

PhanthyCode desktop releases and CLI compatibility notes.

v2.34.0StableLatest
  • PhanthyCode and PhanthyWork 2.34.0 are available for Windows, Apple Silicon, and Intel Mac.
  • Smart execution reduces unnecessary approvals for file reads, directory creation, document opening, and subtask launches. Combined commands are checked individually while preserving granted scope.
  • Improves long-task stability during temporary connection failures. Failed requests can be retried in the same conversation or resumed from a chosen point.
  • Improves conversation recovery when space records are unavailable and preserves recovered tasks’ space membership.
  • Refines main-task and subtask progress, reply timestamps, and scrolling; reduces empty space after completion and hides final-reply actions on intermediate progress messages.
  • Improves HTML preview scrolling and the in-app browser opening experience. Empty-response errors remain visible after history refresh for clearer recovery.
  • The student area now focuses on experts, with its dedicated skills section removed. Daily reward dismissal is clearer.
Current version downloads
macOS · Apple SiliconmacOS · Intel⊞Windows · x64🐧Linux · offline package (tar.gz)
v2.33.2
  • Reduces flicker, scroll jumps, and position changes during streaming replies and at completion.
  • Pauses automatic following when you scroll or interact; use Latest or send a message to resume.
  • Preserves composer height, drafts, and viewing position when switching between main conversations and subtasks.
  • Expands available thinking before the answer starts, then collapses it once while respecting manual choices.
  • Refreshes launch-screen branding and improves expert and skill category navigation.
v2.33.1
  • Fixes Auto reporting an unavailable model and blocking messages, and improves selection among models available to your account.
  • Balances task complexity, speed, and cost. Routing can be cancelled, and retries reuse the selection to avoid duplicate routing charges.
  • Includes settled model-selection charges in the current task and conversation history.
v2.33.0
  • PhanthyCode and PhanthyWork 2.33.0 are available for Windows, Apple Silicon, and Intel Mac.
  • Adds /goal for persistent objectives with progress, pause, resume, and cancellation. Rounds and duration are unlimited by default; goals survive an app restart and resume when you choose.
  • Browse read-only subtask conversations, tool calls, execution status, and settled credits, then return to the main conversation to continue.
  • Improves per-turn credit totals, including associated subtasks and saved history. Missing charges are hidden instead of appearing as a misleading zero.
  • New users start with Auto. After a conversation, new tasks remember the last model actually used, while historical conversations restore their own model selection.
  • Simplifies sidebar and task status indicators. Completion time and duration appear at the end of the reply footer on hover, with less layout movement when replies arrive or history is reopened.
  • Improves recovery guidance for task and space records. Incomplete conversation history keeps a specific error and a recovery option that preserves the original conversation and files, reducing ineffective retries.
v2.32.2
  • Preserves independent tasks and space membership across navigation, refreshes, and restarts, including empty spaces.
  • Move tasks, rename spaces, and remove spaces without changing working directories or deleting conversations and local files.
  • Fixes incorrect space names, incomplete directory labels, and task membership changes after completion.
  • New tasks start with Auto while historical conversations keep their own model selection; later releases further refine model preferences.
  • Keeps other channels connected when binding or reconnecting WeChat and preserves existing channel configuration.
v2.32.1
  • Fixes duplicate launches after macOS updates by waiting for the previous process to exit.
  • Prevents concurrent instances of the same product while keeping PhanthyCode and PhanthyWork independent.
  • If restart fails after installation, retry restarting without installing again.
v2.32.0
  • Rate completed results with reversible likes and dislikes. Add reasons and comments, and preview or exclude attached conversation content.
  • Shows settled credits and input/output usage details, preserving feedback actions and usage records in conversation history.
  • Reduces repeated balance refreshes across account, model, and plan views while updating after rewards arrive.
  • Uses a clear Install and Restart action after confirmation, reducing indefinite restart waits.
v2.31.0Beta
  • 2.31.0 is being prepared for testing; release installers are not yet available. This page continues to offer the stable 2.30.0 downloads.
  • Expert and skill menus add search, descriptions, and management shortcuts, including local skill import directly from the composer.
  • Dropped attachments show processing status and can be cancelled. Sending waits until attachments are ready to help prevent missing files.
  • Output lists hide files that no longer exist. After a task finishes, successful late-arriving execution records move into process details to keep the final reply clear.
  • Removes the extra task-turn limit imposed by Desktop and warnings triggered solely by long-running steps, while retaining recovery prompts for actual connection failures.
v2.30.0
  • PhanthyCode and PhanthyWork 2.30.0 are available for Windows, Apple Silicon, and Intel Mac.
  • Improves @ file and folder references, fixing path-search errors and stale results replacing newer input.
  • Question cards are more compact, single-choice answers advance to the next question, and bottom controls support navigation and custom-answer submission.
  • Fixes missing automatic HTML previews and blank previews after clicking. Newly generated HTML opens automatically without repeatedly taking over the preview when revisiting history.
  • Improves question and approval recovery in the experimental engine, reduces repeated activity, referral, notification, and model-list refreshes, and prevents stale data after switching accounts.
v2.29.0Coming soon
  • 2.29.0 preview: this update focuses on sign-in recovery, clearer model errors, and more reliable task execution. Installers are being prepared.
  • Adds fallback sign-in options to open or copy the login link and enter an authorization code after browser approval. Sign-in can be postponed so its dialog does not lock the entire app.
  • Distinguishes model permissions, service quotas, rate limits, and image support errors, with clearer next steps to reduce false sign-in warnings and ineffective retries.
  • Improves Git environment detection and recovery after installation on Windows to reduce repeated installation prompts.
  • Corrects how credit usage is attributed to conversations, expert execution permissions, and tool status displays so task progress better reflects execution.
v2.26.1
  • 2.26.1 is available for Windows, Apple Silicon, and Intel Mac, combining 2.26.0 improvements with follow-up fixes.
  • Adds a student area with learning and research expert categories. Verification cards show actual status and awarded credits, with sharing incentives available after verification.
  • Improves Power Plan task ordering, permission recovery, expert installation retries, built-in engine support for experts, and skill categories.
  • Previews local HTML and relative assets inside the app, with consistent path handling across chat, the address bar, and workspaces, including spaces, Chinese characters, and special characters.
  • Validates real preview paths to prevent out-of-scope symbolic links from exposing unauthorized files. Windows research workflows retain required interactive confirmations.
v2.26.0
  • Splits WeChat replies at sentence and paragraph boundaries with a timeout buffer, reducing truncated responses and broken links.
  • Shows context before tool execution or approval, retains partial results on failure, and stops pending WeChat output when a task is stopped in Desktop.
  • Restores messages from externally initiated conversations without replacing an active new turn during history refresh.
  • Repairs Windows browser login callbacks and surfaces delayed login links. Cancelled and retried requests can no longer overwrite the current login state.
v2.25.0
  • Fixes skill selection failing to insert commands on the New Task page, with better keyboard selection and draft preservation.
  • Adds a 3 MB local image limit and built-in compression to control model context costs. Failed compression no longer lets oversized originals through.
  • Fixes rejected Windows working directories and duplicate session occupancy, including paths with repeated separators or trailing slashes.
  • Preserves original tool inputs and reuses approved directory scope within a session. Approval cards show submission progress and retryable failures.
  • Restores pagination and root-level documents in the Feishu picker, with better initial loading, search, and deduplication.
v2.24.3
  • Reworked conversational questions and approvals so pending input from both main tasks and subtasks remains visible.
  • Restores unanswered interactions after reconnecting, refreshing, or switching windows, and keeps failed submissions available for retry.
  • Queues multiple pending requests in order and narrows shortcut scope to reduce conflicts with input methods and system shortcuts.
  • Moves large-conversation guidance into the relevant chat and improves marketplace access, project separation, and loading feedback.
v2.24.2
  • Fixes multi-step choices and recommended answers that could not be submitted by matching pending questions reliably.
  • Keeps accessible Feishu documents available and adds clear recovery guidance when the picker cannot load.
  • Hides empty sessions with no messages and adds scoped undo and redo shortcuts across common editors on macOS and Windows.
v2.24.1
  • Adds Repair and retry now for Windows startup failures, recovering runtime state without deleting user content or configuration.
  • Adds one-click diagnostic export to Downloads for easier support follow-up.
  • Keeps PhanthyCode and PhanthyWork diagnostic filenames and data scopes separate.
v2.24.0
  • Improves Windows recovery after unexpected exits and under system resource pressure, reducing restart loops and unresponsive states.
  • Makes post-update relaunch more reliable and displays release notes in the application language.
  • Adds Copy to the selected-text menu and lets users answer a single pending question directly from the composer.
  • Clarifies feature entry points between PhanthyCode and PhanthyWork so each product stays focused on its intended workflow.
v2.23.2
  • Keeps context-compaction progress and elapsed time visible, with clearer feedback when the operation is unusually slow.
  • Restores missed confirmations and approvals automatically while removing stale connection warnings after completion.
  • Gives long tasks more room to finish while reserving capacity for a useful final written handoff.
  • Opens generated Office, PDF, and archive files reliably from the verified workspace and prevents reminder cards from overlapping.
v2.23.1
  • Restores the intended public branding, launch experience, service configuration, and update behavior for PhanthyCode and PhanthyWork.
  • Expands the sidebar account menu with direct access to plan, referral, appearance, update, help, and sign-out actions.
  • Fixes signed-out account labels and avoids resending an unchanged image in the same context, reducing unintended usage.
v2.23.0
  • Simplifies the main model picker by removing distracting internal tier labels while retaining useful details in the full catalog.
  • Uses Enter to send and Shift+Enter for a new line on new profiles while preserving existing shortcut preferences.
  • Streamlines completion metadata while keeping the model, input and output usage, and completion time visible.
v2.22.1
  • Adds file drag-and-drop and clipboard paste to the composer, with direct preview, open, and reveal actions for generated results.
  • Keeps long-running task status visible through actual completion and adds subtask progress, waiting states, and completion time.
  • Recovers from startup network failures and prevents slow integrations from blocking the first conversation for an extended period.
  • Improves long-conversation reliability, generated-file opening, and update prompts with clearer continue and retry actions.
v2.22.0
  • Restored Light, Dark, and System appearance modes; the preference is preserved, System mode reacts immediately to operating-system appearance changes, and home guidance, action hints, and message labels now retain clear contrast in Dark mode.
  • Added a conversation-history entry and full-conversation search to the chat header, with direct navigation to earlier questions and matching answers; counts above 99 stay compact as 99+.
  • Redesigned scroll timing across send, streaming, and session switches so new turns settle at the latest content, manual reading is respected, and long responses produce substantially less viewport jumping.
  • Stabilized the streaming status card and added the active tool and completed step count, while completion metadata now presents the model with its product mark and name.
  • Fixed generated files taking over the app window when opened or downloaded; HTML, documents, spreadsheets, presentations, and archives can use the system default app, and saving a copy no longer replaces the current page.
  • Preserved the 1M context capability for long-context models and corrected overlapping cached/input token accounting so a single large task is not compacted early because usage was overestimated.
  • Simplified composer readiness: real input enables the button and Enter whenever a submission is not already in flight, while model, runtime, and history recovery problems now provide explicit feedback instead of leaving the control permanently gray.
  • Restored Scheduled Tasks as a top-level entry below IM Channels; successful creates and edits now close immediately while notification setup finishes in the background, and short task lists scroll naturally with the page.
  • Resequenced daily startup notices so the first launch prioritizes the daily reward, the second can surface an unfinished Power Plan, and users who completed every task are no longer prompted.
v2.21.0
  • Power Plan now follows the currently open campaign round, keeping Round 3 enrollment, participation status, and stage guidance current in the desktop app instead of relying on fixed Round 2 copy.
  • Model context windows now combine live catalog configuration with each model's actual capacity, allowing Phanthy and other long-context models to use their available window and reducing premature context compaction.
  • Added consecutive and per-turn safeguards to automatic compaction, preventing repeated “Context automatically compacted” loops when a model or service fails while preserving a clear error for recovery.
  • Windows now detects and attempts to recover from an unexpectedly stopped background service or a prolonged unresponsive state, reducing cases where the entire app appears to ignore clicks until it is restarted.
  • Older Windows processors that lack the required instruction set now receive a clear hardware compatibility explanation before startup instead of repeated background-service crashes after installation.
  • Refreshed the product mark used by Phanthy models for more consistent branding in both the model list and the selected-model control.
v2.20.4
  • Added Phanthy, GLM, DeepSeek, and Kimi product marks to the model picker, retained the letter fallback for other models, and now shows the selected model as its mark and name.
  • Recommendation badges now come from the live model catalog, allowing model recommendations to change without another desktop release while remaining compatible with older catalog services.
  • Moved Contact Support beside Settings with a quieter default treatment and clearer hover feedback, and removed the stale fixed version copy from Power Plan.
  • Windows now removes stale temporary installer directories after update attempts so old update files no longer accumulate on disk.
  • Fixed delayed scroll-follow work from a previous conversation moving the reading position after switching sessions.
v2.19.0
  • Consolidated Experts, Skills, and MCP into one top-level marketplace before IM Channels, with dedicated Expert, Skill, and MCP tabs.
  • Removed the obsolete Agents and MCP pages from Settings so specialist workflows and external services are managed from one marketplace.
  • Desktop-managed MCP services now reach real conversations, including project-private, user, and local services; plugin-provided MCP entries open the matching tab directly.
  • Full auto remains fully automatic for Expert sessions instead of asking again or switching to Smart execution, while tools outside the Expert policy remain blocked.
  • Fixed Windows startup races that could launch multiple Git Bash installers, with stronger installer verification and isolated temporary files.
  • Diagnostic exports now include background-process exit codes, bounded resource trends, Windows commit and page-file snapshots, and recent crash or resource-exhaustion events for idle-time failure analysis.
v2.18.0
  • Requests for additional information now appear directly instead of being preceded by an unrelated permission prompt, including in the strictest permission mode.
  • Generated files retain their complete relative directories, so images, documents, and other results in nested folders open from their real project location.
  • The results area shows more generated images while preserving original relative paths for opening and downloading.
  • Conversations sharing a non-Git workspace now keep separate artifact directories, reducing clutter during long tasks without changing edits to existing project files.
  • Raised the minimum supported PhanthyCode CLI version to 0.3.25.
v2.17.0
  • Points Tasks is now the more prominent Power Plan, with new sidebar shortcuts and a richer task page; after all five tasks are complete, their progress and rewards remain available lower on the page.
  • Refreshed the PhanthyWork brand with a new blue 3D mark, chat avatar, and launch screen, while keeping product names, versions, and guidance such as “You have the idea. Phanthy finds the way.” consistent.
  • Auto edit is now the recommended Smart execution mode: reads, searches, file changes, and safe commands flow automatically while risky actions still ask; Plan mode can continuously read and analyze without gaining edit access.
  • Allow for this conversation now follows the session and is reused on later turns, reducing repeated confirmations; read-only analysis of large slide decks or document sets no longer prompts once per file.
  • Improved customer-service and community image loading, retries, and production-domain compatibility, with clearer loading and failure states for remote images.
  • Fixed retry loops caused by terminal control messages and authentication failures, duplicate cards for the same error, and the interface switching to English after phone registration in Chinese.
v2.16.0
  • PhanthyWork now includes Points Tasks for eligible new and existing users, awarding points for real conversations and Expert or Skill experiences.
  • Added an independent experience-sharing reward entry with multiple submissions, per-post review status, and rewards that can stack with referral points.
  • Points notices can now stack, task progress refreshes automatically, and the five-task progress section collapses after completion.
  • The Expert Market now has a loading experience and short-lived cache, and correctly handles experts whose installed version is out of date.
  • Fixed the Thinking Strength submenu pointer path and restored the Skill menu when a task opens the composer with a slash.
  • Improved sign-in authorization, Windows runtime stability, long-task waiting and service-error recovery, with previews for images returned by tools.
v2.15.0
  • Fixed sign-in failures on some Windows devices by detecting stale or non-runnable CLIs in the user profile and repairing the runtime automatically, even after a desktop reinstall.
  • Restored the close, minimize, and maximize controls in the top-right corner of PhanthyCode windows on Windows.
  • Corrected macOS conversation layout and scroll boundaries to prevent an extra scrollbar above the composer and unbounded downward scrolling.
  • Gateway failures now use bounded retries and circuit breakers instead of retrying indefinitely, while preserving a clear failure state for diagnosis.
  • The desktop app now shows a bottom-left notice when referral or admin-granted points arrive, with direct links to referral history or account details.
  • Introduced a new referral ladder for newly qualified referrals: 5,000 points for the first, 8,000 for the second, and 10,000 for each referral from the third onward.
v2.12.0
  • Added progressive daily start rewards that rise from 300 to 1,500 points, resetting to day one after a missed day.
  • The daily reward notice now shows the streak day and actual server-granted points, while older clients remain compatible.
  • Fixed an issue where completed task lists could continue processing and prevent the final result from appearing.
v2.10.0
  • Improved first-launch and post-login reliability, reducing cases where a new conversation could not start after setup or an account change.
  • Made model selection and message details clearer, with easier access to common models and per-message usage information.
  • Unified the Contact Us entry so you can scan to reach support or join the user community; scheduled tasks now live in Settings.
  • Improved update reliability and everyday desktop interactions.
v2.9.0
  • The composer + menu now includes installed Skills, shown with their marketplace display names and inserted as the matching command.
  • Expert and Skill flyouts now switch cleanly without overlapping, and Skills are available from both new tasks and conversations.
  • After switching a project directory, the first message can be sent as soon as repository validation completes on macOS, Windows, and Linux.
  • The local OAuth callback now returns a success page before attempting to close itself, preventing a localhost error after successful authorization.
  • Dark mode has clearer text, canvas, sidebar, panel, and divider contrast; active tasks show concise progress with the full process available after completion.
  • Improved Feishu connection and document authorization reliability, and unified desktop version metadata at 2.9.0.
v2.8.0
  • PhanthyWork Beta and PhanthyCode now ship as separate macOS installers: Work uses the blue Beta brand while Code keeps the monochrome stable brand.
  • The two desktop apps use separate bundle IDs, local configuration directories, install names, and update channels, so builds with the same version never cross-update.
  • Fixed PhanthyWork's default window dimensions and macOS title-bar behavior, restoring the intended immersive application window.
  • PhanthyCode download actions on the homepage now lead to the product download page; PhanthyWork keeps quick download, with every Work picker sharing one component and copy source.
  • Refined Work wording across product, download, and documentation pages to distinguish the PhanthyWork AI Agent workspace (Beta) from the PhanthyCode coding assistant.
v2.7.0
  • Added a Skills marketplace for discovering curated capabilities, understanding when to use them, and applying installed Skills directly in conversations and tasks.
  • Expanded the Expert marketplace across office, research, data, product, and engineering scenarios, with clearer introductions and starter questions.
  • Unified Expert and Skill discovery, installation, and launch flows so installed capabilities remain recognizable and new tasks start faster.
  • Improved response structure and file delivery for complex work, making key conclusions, progress, and generated results easier to follow.
  • Improved desktop reliability and everyday polish, including startup and task-completion reminders plus a range of interaction fixes.
v2.6.0
  • Added an Expert marketplace with expert details, strengths, starter prompts, and direct expert handoff into a new task.
  • Connected expert installation to the real runtime: remote packages are checked for origin, hash, size, and archive safety before PhanthyCode CLI loads them in conversations.
  • Kept the Skills entry as a coming-soon preview; Skill marketplace installation and enable or disable controls are still in preparation and are not released in 2.6.0.
  • Upgraded the website /skills route into an Expert catalog with detail dialogs and a PhanthyWork download handoff for web summons.
  • Kept home shortcuts as prompt templates without presenting a shortcut as a released Skill.
  • Improved activity exposure so the home notice remains visible until explicitly dismissed for the day, and consolidated the Beta and version labels into a lighter badge.
v2.5.0
  • Upgraded the desktop product to PhanthyWork across the app name, window title, Dock label, system notifications, and installer presentation while preserving internal identifiers and update compatibility.
  • Refined the desktop sidebar brand area so the full PhanthyWork name fits when expanded and the original icon remains correctly proportioned when collapsed.
  • Updated desktop-workspace entry points, Token Plan, download guidance, and brand copy across the website, while unifying button, icon, and page colors.
  • Kept PhanthyCode as the command-line tool name, with CLI commands, documentation paths, and existing links unchanged.
v2.4.1
  • Improved long-running task visibility: desktop heartbeats now report the CLI process, communication bridge, latest progress, and pending permissions, with a clear warning after three minutes without progress.
  • When the CLI or bridge is confirmed disconnected, the desktop now attempts one recovery of the original session without replaying the previous potentially side-effecting command; manual Reconnect and Continue remains available.
  • Fixed concurrent permission prompts overwriting each other and causing Tool permission stream closed, while restoring unresolved permission requests after reconnection.
  • Fixed empty pages values being rejected by the Read tool, and permission-mode changes incorrectly showing a thinking state and Stop button while idle.
  • Restored sent-image thumbnails and full previews after history reloads, while hiding internal image-source metadata added automatically by the CLI.
  • Added recovery requests, recovery results, and connection-health snapshots to local diagnostic logs so model, CLI execution, and desktop communication failures are easier to distinguish.
v2.4.0
  • Moved the app to the PhanthyCode-owned bundle identifier com.phanthy.code.desktop, with migration for portable-mode, terminal, and window-state settings stored under the legacy identifier.
  • Refreshed the app icon with a clearer blue-green gradient and a larger, brighter white PhanthyCode mark.
  • Kept the existing v2 update channel and pinned updater key so desktop versions already on v2 can follow the established upgrade path.
  • Refined the Service settings hierarchy by placing sign-in details first and separating the app and CLI engine version sections.
v2.2.1
  • Added a daily start-of-work reward for eligible users, with campaign balances and reward history available from the Token Usage page.
  • Improved in-app updates: recommended releases now remind first and download only after confirmation, while required releases prepare automatically and block new tasks until updated.
  • Added clearer retry, official-download, and temporary-continue options when an update fails, so network issues do not interrupt existing work.
v2.1.0
  • Added the PhanthyWork Token Factory flow, with verified-phone enrollment, desktop activation, eligibility status, reward delivery, and one clear next action.
  • Upgraded the points wallet so plan points and campaign rewards are recorded separately while working across the same public paid models, with total, plan, and reward balances shown explicitly.
  • Connected reward delivery to real product use: start a first task after credits arrive, then continue the most recent work after the first successful model request.
  • Improved plan and model views so users without a paid subscription can still understand reward-funded access, available models, and balance sources.
  • Fixed monthly and daily plan usage displays so one points pool is no longer mislabeled as separate 5-hour and 7-day allowances, with additional composer and campaign-page layout refinements.
v2.0.0
  • Upgraded model selection from the fixed Fast, Standard, and Pro tiers to a server-driven catalog with model capabilities and billing multipliers.
  • Added reasoning-effort controls for supported GPT and GLM models, with the active model and selected effort kept visible in conversations.
  • Improved catalog caching and failure fallback so existing choices remain stable and a reachable free model is preferred when the service is temporarily unavailable.
  • Admin conversation logs now show the reasoning effort actually sent with each request for easier parameter verification.
  • The homepage Download Now action now opens a quick installer picker, starts the selected download, and continues to the download page for installation notes.
v1.9.0
  • Introduced the Expert Market with installable specialists for research, business documents, presentations, data analysis, project management, meetings, finance, executive assistance, SOPs, file organization, sales proposals, customer success, and recruiting.
  • Experts can be selected for a new conversation or switched during an existing one, with the active identity shown explicitly and its Agent instructions and Skills loaded into the session.
  • Added network discovery through domestic MCP and Skills hubs, kept clearly separate from the official catalog; external results can be handed to a new ask-permission Phanthy draft for review, installation, and verification in the local catalog.
  • Improved expert installation with confirmation, visible progress, and retry feedback, while fixing newly installed experts not appearing promptly in the conversation selector.
v1.7.0
  • Refined the desktop layout, sidebar spacing, title bar, and new-session composer for a cleaner daily workflow.
  • Added clearer hover and selected states across sidebar navigation and session tabs; new drafts now enter the list after the first message, while composer context controls stay simpler.
  • Smoothed long conversation history scrolling by pre-rendering nearby messages and preserving the reader's position.
  • Fixed half-size window restoration on Retina Intel Macs, refined macOS window-control placement, and refreshed the signed installers.
v1.6.0
  • Improved session permission prompt handling so mobile IM approvals stay aligned with the active desktop session.
  • Enhanced CLI auto-install diagnostics so setup failures are easier to identify and recover from.
  • Includes Windows CLI install, update, and detection fixes for a more reliable first-run desktop setup.
v1.5.0
  • Improved diagnostics export and startup-failure recovery with direct access to PhanthyCode diagnostic bundles.
  • Expanded dev telemetry probes across RUM and product events for login, update, and chat failure verification.
  • Documented the supervised release flow and smoke-test checklist while keeping macOS and Windows downloads on stable latest links.
v1.4.0
  • Improved desktop observability for login, update, and chat failures.
  • Enhanced product-event reporting and admin trace-back for retention analysis.
  • Updated macOS and Windows installers while keeping latest download links stable.
v1.3.0
  • Added an anonymous diagnostics toggle for basic stability and error analysis.
  • Usage records now separate source and OS, making desktop and CLI activity easier to read.
  • Improved IM bot setup, pairing-code copy, and product wording.
v1.2.0
  • Fixed an error during forced update or CLI install that could block setup.
  • Fixed scheduled-task reminders that did not fire reliably at their due time.
  • Further stability improvements for resuming chats, the usage page, and long-running tasks.
v1.1.1
  • A maintenance patch improving stability for older chats, usage pages, and long-running tasks.
  • Fixed some chats failing to continue and occasional message display issues.
  • Usage and credits pages now show account status more reliably.
  • Large chats and long-running tasks now provide clearer feedback and smoother responses.
  • Windows install and update logs are cleaner with fewer unrelated notices.
v1.0.2
  • Our first stable 1.0 release — works out of the box: sign in and start chatting for free.
  • Steadier and smoother: chat, sign-in and install are more polished, with fewer hiccups and faster recovery.
  • New update notifications: a gentle prompt keeps you on the latest version.
v0.9.0
  • A maintenance release fixing several login, update, and model-selection issues.
  • Fixed being able to keep chatting after signing out.
  • Fixed occasional login failures that required retrying.
  • Fixed Update not reaching the latest version.
  • Fixed the wrong default model for new chats and other models leaking into the picker.
  • Fixed chats getting stuck after an error — they now recover automatically.
  • Windows: fixed several install, update, and startup issues.
v0.8.0
  • A stability & polish release focused on fixing several chat, connection, and CLI issues.
  • Fixed multi-turn chats that could hang or cross-talk under rapid sends or multiple windows.
  • Fixed long replies occasionally stuck on "Thinking" and permission prompts not appearing.
  • Fixed new messages being interrupted right after tapping "Stop".
  • Fixed sessions occasionally becoming unresponsive after reconnecting.
  • Fixed sign-in status mismatches and detection / startup failures in some environments.
  • Fixed occasional scheduled-task startup failures, plus other stability and detail improvements.
v0.7.0
  • Subscriptions & plans: models unlock by plan — paid models are greyed out with a clear upgrade path when you're not subscribed, the free tier works out of the box, and new accounts are guided to subscribe after signing in.
  • Refreshed look: a neutral color palette and a new app icon, plus Light / Soft-dark themes and a tighter, more consistent UI.
  • Menu-bar quick menu: click the status-bar icon to start a new chat or open Settings.
  • Usage page polish: a nicer loading state, more accurate sign-in status (with retry and sign-out), and clear guidance to subscribe when you don't have a plan yet.
  • Install & compatibility: better detection of an already-installed CLI (more install methods supported), with guided install when it's missing.
  • More details: model pricing info, tap to view the current conversation's context usage, and clearer retry hints.
v0.6.0
  • Usage center upgrade: see your plan's credit balance, usage broken down by day and model, and recent requests on the Usage page — consistent with your account page.
  • Model pricing in the picker: each model now shows its per-credit rate when you switch.
  • Context usage at a glance: click the context button to instantly view the current conversation's context usage — faster and lighter.
  • UI & copy polish: consistent wording and model names, plus fixes to the tool-call copy button and message layout.
  • Stability improvements: better session list, permission memory, and context loading behavior.
v0.5.0
  • Model picker unified to Phanthy plans (Standard / Fast / Pro); models from other sources are no longer mixed in.
  • macOS signing & notarization restored — opens right after download, with no security warnings.
  • Fixed the CLI failing to start in some environments.
v0.3.0
  • Earlier stable release.
PhanthyCode 更新日志 | PhanthyCode