Plans and pricing are now available inside RevampKeyreview available plans and start checkout without leaving the app.
Manual model selection is now available through paid planstrial accounts are guided to upgrade when they choose it.
Changed1
Concurrent work targeting the same repository is coordinated more safelyGit and file changes no longer interfere with each other when multiple tasks run at once.
Fixed1
RevampKey now starts correctly on systems without Git installedno-Git workspaces no longer fail during startup.
Interactive Mermaid diagrams now open directly in chatexplore visual flows without leaving the conversation, with clearer cards and theme-aware colors.
PDF attachments are supported alongside imagesshare more of the material you are working with in a request.
Subagent activity is easier to follow as work happensgrouped runs, live status, clearer activity titles, and visible results make background work easier to understand.
Concurrent tool and AI work can run in parallel more reliablylonger requests spend less time waiting on independent work.
MCP workflows provide broader guidance for connected serviceswork with more provider capabilities and setup paths.
Changed5
Attachment handling is more dependabletransfers are acknowledged correctly, failed uploads preserve your draft, and attachments stay with the request that supplied them.
Diagrams render more consistently across chat sizes and can be saved through the appvisual content remains usable from creation through export.
Plan conversations are ordered by their latest activitythe work you touched most recently is easier to find.
Search-and-replace results explain multi-edit changes more clearlypreviews better match the changes that will be applied.
Auxiliary sidebar defaults are balanced againopening the workspace starts from a more predictable layout.
Fixed4
Confirmation prompts no longer compete with each otherconcurrent requests wait their turn instead of presenting overlapping choices.
Repository-map context is no longer sent to code subagents when it is disableddelegated investigations follow the selected workspace-context settings.
The obsolete Source Control commit-message action has been removedthe SCM view no longer shows an unavailable command.
Interactive diagrams no longer appear with incorrect viewport dimensionsflowcharts remain visible and proportioned correctly in chat.
Live plan usage is now visible and updates after each completed requestsee how much of your plan capacity you have used, including public-rate plans.
Earlier summary records can be opened to recover conversation contextfind the details you need when returning to long-running work.
Notebook edits now show a structured preview before changes are appliedreview source changes and cleared outputs before opening the notebook.
Notebook and individual cell contents now appear directly in chatinspect notebook results without leaving the conversation.
The Agent navigation rail can be resizedgive active work more room while keeping the workspace comfortable to scan.
MCP-aware workflows now explain configured-but-disabled servers and available connector optionsunderstand what needs credentials or activation before continuing.
Code subagents can safely inspect shell and runtime behaviorinvestigate projects that need more than static file discovery.
MCP operations now use scoped delegation, provider-specific guidance, and current tool schemaswork with connected services more reliably.
Changed9
Related tool activity and recovered conversation history are grouped more clearlycompact reports and replies are easier to scan.
Approval and prompt cards now wait for your explicit choicethey no longer select a default after a timeout.
Conversation compaction preserves more of your work across restarts and checkpoint revertsuser decisions, complete turns, reasoning, tool results, tracker context, revisions, and subagent edits remain recoverable.
Search and environment discovery cover more workspace paths, project types, and local toolsrepository investigation is broader while remaining bounded.
Git workflows preserve structured targets and ranges, support bounded retries, and keep oversized terminal output recoverableimportant context is less likely to be lost during operations.
Delegated research now carries the current request, reuses reports already available, and limits automatic handoffs to the active reflectionfollow-up work stays focused and avoids unnecessary repetition.
Planning and delegated work stay aligned with the current project statetracker and specification context remains authoritative as work progresses.
Billing and request metadata remain available through completion and persistenceusage-related responses continue to carry the information needed by the app.
MCP operation reports are easier to find after delegated work completesthey are grouped with other operations instead of being misplaced.
Fixed13
Failed requests, prompt errors, and unexpected disconnects now remain visibly interrupted and retryableinstead of appearing complete.
Plan and tracker updates now route to the root plan correctly after a cold start or child task session.
The Revamp AI layout-control icon is available again in Editor mode, while maximized auxiliary views hide redundant layout actions.
Stale subagent calls from ended sessions are marked abandonedinstead of remaining indefinitely pending.
Failed streamed tool calls are no longer dispatched as if they were valid, and interrupted operations report an unknown outcome instead of a misleading failure or success.
Prompt construction and caching no longer hit the identified crash paths.
Completed sibling tool results are retained when another tool in the same batch is slow.
Conversation summaries and checkpoint reverts no longer silently lose content, surviving segments, user decisions, or subagent edits.
Search-and-replace accepts unambiguous partial-line matches and explains the actual mismatch when a replacement cannot be applied.
Stopped subagents retain useful work and handoff contextinstead of collapsing to a single interruption notice.
Attachments remain scoped to the message that supplied theminstead of leaking into later work.
Public-rate billing rejections now explain the server-provided reason and avoid futile retries.
Large repository maps avoid dependency trees and remain bounded on unusually large workspaces.
Artifact search and retrieval results now appear directly in chatinspect past work and supporting evidence without leaving the conversation.
Past Work Reports settings give you control over artifact searchchoose whether the main agent and subagents search the current conversation, specifications, workspace, or all available reports.
Codebase structure sharing can now be configured independentlykeep the context available to the main agent and subagents aligned with the way you work.
Changed4
Long-running AI work is easier to resumeinterrupted subagent work retains more of its context and can continue after a restart instead of requiring a fresh run.
Research context is easier to carry forwardearlier reports, repository identity, attachments, and follow-up context remain available when later work builds on them.
Queued follow-ups now preserve the original request while applying consistent safety checksthe message you entered remains the message the assistant processes.
Agent navigation spacing is more consistentthe workspace rail is easier to scan while moving between active work.
Fixed6
Queued follow-ups no longer enter the active run too earlythey wait for the correct handoff point before being processed.
Queued messages now replay consistently after a restartexact model-facing content is preserved instead of being regenerated differently.
Inline follow-ups resume with the correct conversation contextcontinuing work no longer loses the connector-owned handoff state.
Artifact search and retrieval results render safely and clearlytool responses no longer show redundant metadata or fail to display usable results.
Main-agent and subagent context stays separated correctlycodebase structure and artifact scope no longer leak across agent boundaries.
Attachments and summaries preserve the right order and contextlonger multi-step work is less likely to lose evidence or continue from stale state.
A new agent-first workspace makes active work easier to followprojects, conversations, and tasks are organized in a dedicated navigation rail with nested conversations, processing indicators, timestamps, and quick new-task actions.
Plans now have a dedicated workspacereview modules, live file changes, and specifications in separate tabs, open changed files in the editor, and resize or collapse the plan pane to fit your workflow.
Settings now have a redesigned homeAccount, Usage & Plan, Models, Code Structure, Git Workflow, and MCP Servers are organized in a clearer, responsive layout.
The workspace now remembers more of your layoutexpanded projects and the last active conversation are restored when you return.
Changed4
Agent and Editor mode feel more consistentthe composer, landing screen, navigation, and surrounding controls now share the same polished visual language.
The new workspace is optimized for focused AI workwide layouts use a calmer hero/composer experience, clearer context controls, and less distracting chrome.
Tool activity is easier to scanlow-attention tool calls are grouped into concise summaries, while important file changes and results remain easy to expand.
The default dark theme has a refreshed Graphite and Violet palettenavigation, cards, controls, and focus states have clearer hierarchy and contrast.
Fixed5
Changed files and diffs are more reliablestaged, unstaged, untracked, empty, and large diff states now open and display with the correct scope.
Plan-pane navigation is more dependableselecting a tool result or changed file opens the right destination, including falling back to the editor when no plan view is available.
Sidebar and plan-pane interactions work consistently across layoutsproject rows, conversation selections, collapsed pane tabs, and wide-mode clicks now respond as expected.
The chat workspace no longer reloads unnecessarily on startupcold starts avoid duplicate initialization and preserve the active context more reliably.
Composer states and uploaded images render correctlyworking indicators, image thumbnails, focus borders, and spacing no longer get stuck or drift between modes.
Background AI work is now visible in your conversation listactive conversations show a live indicator even when you are viewing another conversation.
Multi-repository workspaces now provide repository-level Git statusRevampKey can distinguish the status of each repository instead of treating the workspace as one undifferentiated folder.
Searches can include or exclude hidden and gitignored files independentlyfile discovery gives you more control while preserving which results come from ignored files.
Changed10
AI responses are easier to readsource lookup markers, file paths, and line-range tags are no longer shown in chat, while the referenced code and formatting remain intact.
Progress indicators now reflect live activityprogress and stop-button animations appear only while the assistant is actively processing.
Model changes now create a clean conversation boundaryswitching models during a conversation preserves the conversation state more reliably, and notices use the model’s display name.
Workspace context is carried through more reliablyAI requests retain the relevant workspace folders, repository scope, Git context, and shared prompt context across multi-root workspaces and recursive planning.
Terminal results can include more outputlonger command output is available before it is truncated.
Conversation summarization is more reliablesummaries avoid repeated processing, preserve workspace scope, and retain memory across the appropriate project context.
Large-context subagent work is more resilientoversized requests are handled more safely instead of failing because the input is too large.
The desktop layout is easier to navigatepersistent sidebar controls now appear in the title bar, title-bar spacing is improved across platforms, and default sidebar widths adapt better to the window size.
New launches go directly to the workspacethe startup editor is disabled by default.
Dark mode has clearer controls and hierarchycheckbox borders are visible and sidebar titles use a quieter contrast.
Fixed3
Extension browsing is available againthe built-in extension marketplace configuration has been restored.
Explicit auxiliary-sidebar visibility settings are now respectedRevampKey no longer overrides a user’s chosen visibility.
Workspace storage and conversation checkpoints now use the correct locationAI operations are less likely to lose the workspace context needed to restore or continue work.
Queued messages now show clearly while the assistant is busysending a message while the assistant is still replying shows it as queued and picks it up automatically once its turn starts.
Multi-step assistant responses are now saved as individual stepsso your conversation history accurately reflects each action taken, instead of merging everything into one message.
Changed1
Follow-up messages are picked up soonera message sent during an active response is now picked up as soon as the assistant reaches a natural pause, instead of waiting for the entire response to finish.
Fixed9
Queued messages no longer look stuck or out of orderfixed queued messages sometimes appearing out of order, showing a stuck highlight, or losing their status indicator once picked up.
Fixed a rare conversation-corrupting race conditionsending a new message while the assistant was still replying could occasionally corrupt or crash the conversation; this no longer happens.
AI sessions start up more reliablyfixed slow or failed AI session startup when GitHub was unreachable.
Clearer error when a firewall or VPN blocks the connectionSSL/TLS connectivity failures now show a plain Retry/Cancel dialog instead of failing silently after a long retry delay.
Clearer error on full network losswhen your connection drops completely, RevampKey now shows a Retry/Cancel dialog instead of appearing to hang.
Friendlier error when the AI provider has an outagea mid-response provider failure now shows a plain-language message instead of a raw error, and the conversation stays usable.
Fixed new conversations failing on the very first messagean unavailable model no longer gets auto-selected behind the scenes, which was blocking some fresh workspaces from sending their first message.
Fixed the assistant losing context partway through multi-step actionstool results and reasoning no longer get dropped between steps of a longer task.
Fixed an unexpected sign-in browser window appearing when using OpenRouterthe packaged app no longer tries to open an interactive OAuth flow it can't complete.
You can now mention files directly in chattype to search your workspace and attach files as context, with recently used files surfacing first.
Changed1
File mention and attachment chips look cleanerthe picker and chip layout have been refined for a more polished feel.
Fixed4
Model selection and attachments now survive sidebar togglingswitching the sidebar in and out no longer resets your chosen model or in-progress attachments.
Image attachments stay stable when maximizing or minimizing the windowthumbnails no longer resize incorrectly during window state changes.
Fullscreen/maximized view is restored correctly after toggling window modereturning from a mode switch keeps you on the tab you were viewing.
Model changes now reach active conversations reliablyswitching models mid-conversation no longer leaves a stale model applied to ongoing requests.
Model selection now stays aligned more consistently across RevampKeyAuto and Manual mode behavior now follows the same active setting across the app, runtime-backed requests, and follow-up AI work.
Fixed4
Manual model selection no longer loses state while responses are in progressthe toolbar selector now stays stable instead of dropping or resetting your choice during active processing.
Workspace model preferences now persist more reliably between sessionsRevampKey is less likely to drift back to stale model settings after reopening a workspace.
Windows releases now include the expected RevampKey app iconsinstallers and shortcuts now show the branded icon set more consistently across supported sizes.
AI configuration updates now reach the bridge more reliablybackground configuration messages are less likely to fail while RevampKey is syncing AI settings.
RevampKey Icons are now the default product icon themethe editor now uses the updated RevampKey icon set by default, making product branding more consistent across the interface.
RevampKey branding assets have been refreshedupdated logo and icon assets give the app a cleaner and more consistent branded look.
Fixed2
AI bridge startup is more reliablethe bridge now uses a standardized ready signal and a longer startup timeout, reducing failed launches during startup.
Built-in registration no longer uses the wrong marketplace-style flowbundled product integrations now register using the correct built-in behavior instead of an incorrect marketplace-style path.
A new titlebar mode toggle is now built into the custom titlebaryou can switch modes directly from the editor chrome without leaving your current workspace.
Dark mode looks more polished and consistentthe refreshed dark theme colors improve the app’s overall visual clarity.
RevampKey sign-in is now integrated into the appauthentication flows are more seamless and feel more native inside the editor.
Fixed3
AI bridge startup is more resilientRevampKey now falls back to a free local port automatically if the preferred port is already in use.
Sign-in and session recovery are more reliablecallback handling, token refresh, and profile updates are less likely to stall or get lost during authentication.
Error reporting is more dependable when something goes wrongsidebar diagnostics and console failures are captured more consistently, helping problems surface sooner instead of failing silently.
Model dropdown deadlock is goneThe chat toolbar model selector no longer freezes when opened.
AI bridge startup is more resilientRevampKey now falls back to any free local port if the preferred AI bridge port is already in use.
Orphan dialog cards now render consistentlyConversations now show a placeholder AI message when a dialog card arrives without its original assistant message, so the chat no longer appears blank or broken.
RevampAI now starts in Auto model mode by default for new installsNew users start with backend-managed model selection automatically, while manual mode remains available whenever you want to take control.
Model controls are simpler and more compactThe chat toolbar now includes a compact Auto/Manual switch, and settings focus on the primary model instead of exposing internal weak/architect selectors.
Fixed3
Auto/manual model selection now persists reliably across restartChoosing Auto no longer flips back to Manual after restarting RevampKey.
Backend-owned model defaults refresh more reliably for Auto modeAuto-routed requests now follow the backend’s latest main, weak, and architect model defaults without needing stale local selections.
Manual toolbar model dropdown is visible againThe compact toolbar selector no longer gets clipped when opening the model menu.
Strengthened protection of RevampKey's underlying AI engineHardened the app's internals to make them harder to extract or tamper with, with no change to how you use RevampKey.
Required updates can now block outdated app versionsWhen your installed build is no longer supported, RevampKey now shows a dedicated update prompt and guides you to install a newer version before continuing.
Fixed2
Required update notices now show the right minimum versionUpdate prompts now display the supported version details correctly instead of leaving them blank.
Version checks now identify your installed RevampKey build more reliablyCompatibility checks now use the app's actual release version consistently, so supported builds are recognized more accurately.
Chat prompts now start with clearer guidanceThe main prompt box now uses more helpful starter copy, making it easier to describe what you want RevampKey to do.
Tool approval cards are easier to understandConfirmation prompts now hide raw CLI plumbing and show cleaner action wording instead.
Figma help links open more smoothly during MCP setupRevampKey now recognizes Figma's help center as a trusted destination during protected link flows.
Fixed2
Always Allow now actually sticks for approval promptsChoosing Always Allow or Don't ask again now prevents repeated approval interruptions for the same action flow.
Model defaults stay in sync more reliablyModel selection and fallback behavior now update more consistently instead of drifting back to stale defaults.
Custom MCP server secrets are easier to configureYou can now define secret environment variables for your own MCP servers and start from clearer setup examples.
Changed2
Figma connections work with more setupsDesign workflows now support Framelink-based MCP configurations more smoothly across platforms.
Slash-triggered command handling is more predictableTyping a slash now behaves like normal prompt input unless you actually intend to use a command.
Fixed2
Separate AI conversations no longer mix their statePlanning activity, processing state, restored prompts, and interactive cards now stay with the conversation where they belong.
Reply threading is more stable during AI updatesAssistant messages now preserve the correct IDs and stay attached more reliably during streaming and follow-up actions.
New MCP connectors for Filesystem and PostgreSQLYou can now connect to these services directly from RevampKey.
More reliable local connector startupLocal connections now show clearer progress and stay responsive while they launch.
New browser, search, and productivity connectorsBrowser automation, web search, and additional workspace connectors are now available in the assistant.
Changed2
Connector setup and sign-in flows are more consistentSeveral integration flows now complete with clearer handoff and more dependable account linking.
Connector status is easier to followThe assistant now shows clearer activity and success feedback while connections are running.
Fixed3
Interrupted or pending connection flows recover more reliablyStopped or delayed actions now leave the assistant in a better state for retrying.
Tool and response rendering is more stableChat output now stays readable more consistently in compact layouts and mixed-content responses.
OAuth and connector handoff issues are more reliableA number of connection and authorization flows now complete more consistently.
Subagents are more aware of your machine and repository contextRuntime-backed workflows now receive clearer platform details and git identity information during tool-driven tasks.
Changed1
Tool-heavy AI workflows are more environment-awareRuntime context now carries stronger platform and repository details, helping multi-step automation stay aligned with your active setup.
Fixed3
Reflected replies land more reliably in the intended conversationFollow-up responses after reflection now route more consistently.
Runtime-backed replies respect your language more consistentlyAI responses now adhere more reliably to the language you used.
Requests for extra information are better timedThe assistant is clearer about when it genuinely needs more input before continuing.
New plan tracker sidebar for AI planning workYou can now view and follow implementation modules/tasks in a dedicated panel while continuing the conversation.
More flexible RevampAI layoutThe chat and plan areas now support a responsive two-panel experience with a draggable sidebar for better use of screen space.
Built-in codebase structure controls in settingsNew options make it easier to tune how codebase-structure context is generated for AI responses.
Changed3
RevampAI settings are easier to understand and configureLabels and descriptions were rewritten in plain language, with clearer model naming and an improved settings flow.
Chat input is smoother for large contentLarge pasted snippets now collapse into lightweight reference pills so the composer stays readable.
Conversation and plan state now restores more reliablyReturning to RevampAI after hiding/showing the panel better preserves where you left off.
Fixed3
Draft text in chat is more reliableFixed cases where in-progress draft content could be lost during UI transitions.
Narrow layouts now surface planning controls correctlyFixed visibility issues so tracker controls appear as expected in compact panel sizes.
General runtime and packaging reliability improvementsUpdated core runtime integration and build/signing flow for a more stable release experience.
Search is now more reliable when looking through codeImproved fallback behavior reduces missed matches when an initial lookup path fails, so search results are more dependable during everyday work.
Search feels faster and more consistentFile discovery and lookup-heavy workflows now return results with better responsiveness and steadier behavior.
Fixed1
Tool status is now shown correctly after recoveryFixed an issue where search-and-replace status details could appear incorrect after automatic recovery.
RevampAI sidebar now loads every timeFixed an issue where the RevampAI panel would show a blank screen or fail silently on roughly 2 out of 3 cold starts. It now loads correctly and consistently on every launch.
Model picker remembers your selectionThe AI model dropdown was occasionally resetting to the first available model (about 2 in 10 reloads) instead of showing the one you last selected. Your choice is now preserved reliably across sessions.
RevampKey Dark themeA new built-in dark theme with a refined charcoal palette is now included and set as the default dark theme. No extra installation needed; it applies automatically on first launch.
Open Folder shortcut in sidebarWhen RevampKey is launched without a workspace, the sidebar now shows a direct "Open Folder" button so you can get started without hunting through the menus.
Changed4
Faster sign-in experienceThe login screen now appears instantly on first install or after signing out, instead of sitting through a loading animation before being redirected.
AI model display auto-syncsThe selected AI model name in the toolbar now correctly reflects your account's available models right away, without requiring a manual re-selection after sign-in.
Cleaner editor UIRemoved third-party AI assistant icons and suggestions from the title bar, command center, status bar, and Help menu to keep the interface focused and uncluttered.
Smoother extension installsExtension installation no longer shows signature verification dialogs, resulting in a faster and less noisy install flow.
Fixed2
Sign-in retry now works correctlyDismissing the browser sign-in window and clicking "Sign In" again now reliably reopens the browser. Previously, cancelling could leave the login flow in a stuck state.
Cold start "Open a folder" promptFixed an issue where launching RevampKey with no workspace open would show "Initializing Revamp AI…" indefinitely instead of the expected "Open a folder to get started" prompt.
Expanded observability in the integrated RevampKey Runtime with analytics sampling controls, richer message/repo telemetry, and version-aware request metadata.
Added public observability documentation coverage for Loki logging and PostHog event flow to improve operational clarity.
Changed3
Updated integrated runtime from `revampkey-runtime-v0.1.19` to `revampkey-runtime-v0.1.23`.
Improved runtime traceability and analytics identity handling with stronger user/context propagation and extension/runtime build metadata alignment.
Advanced bridge-level and model-request observability context so diagnostics are more actionable across extension and runtime boundaries.
Fixed1
Applied cumulative runtime fixes from `v0.1.20` to `v0.1.23`, including cross-platform compatibility hardening and repo-map crash prevention for empty graph scenarios.
Security1
Continued local-first and privacy-focused defaults with telemetry/analytics disabled by default unless explicitly configured.
Notes2
RevampKey Runtime is the main engine behind RevampKey.
This release carries cumulative runtime hardening and observability maturation through `v0.1.20` to `v0.1.23`, improving reliability, debuggability, and release confidence.
RevampKey Runtime evolution summary integrated into the product changelog to improve transparency around core engine maturity.
Runtime capabilities added since `revampkey-runtime-v0.1.3` include:
Turn-based conversation summarization and improved tool result management.
File memory management/status tracking and stronger context-handling workflows.
Safety checks and cleanup for orphaned/incomplete tool results before API submission.
RevampKey Model Info API integration with local caching, diagnostics, and environment metadata merge support.
Multi-conversation plan architecture with shared tracker/spec resources and restoration/self-healing flows.
Context-management, error-recovery, and design-document prompt systems for more resilient operation.
Prompt compression analysis/utilities and repo map regeneration documentation.
Changed7
Updated integrated runtime from `revampkey-runtime-v0.1.3` to `revampkey-runtime-v0.1.17`.
Runtime and engine-level improvements delivered across `v0.1.4` to `v0.1.17`:
Prompt, cache, and context optimization to reduce token usage and improve responsiveness.
Improved model/provider handling, metadata processing, and API settings behavior.
Repo map generation, timeout behavior, and cache handling refinements.
Dependency and compatibility updates (including tree-sitter/litellm-related updates).
Logging, storage, and conversation lifecycle refinements for better reliability and traceability.
Fixed6
Core runtime fixes delivered across the upgraded range include:
Removal/cleanup of orphaned and incomplete tool-result sequences to prevent malformed message flows.
Cache-key and cache-control reliability issues affecting consistency/performance.
Model/provider routing and metadata edge cases.
Repo map and signature-related reliability issues.
Tracker patch update behavior improvements to preserve existing task/module titles during partial updates.
Security1
Continued local-first and privacy-focused operation with telemetry/analytics disabled by default.
Removed1
Runtime cleanup and maintenance in the upgraded range removed obsolete/unused references and outdated documentation where applicable to keep the engine lean and maintainable.
Notes2
RevampKey Runtime is the main engine behind RevampKey.
This release includes cumulative runtime hardening from `v0.1.4` through `v0.1.17`, with focused work on resiliency, context integrity, cache stability, and tracker correctness.