Repository navigation
Releases
Vulcan Code 0.1.28
Vulcan Code 0.1.28 adds dedicated workspaces and call history for external MCP clients and updates managed ROOT skills at startup when they fall below their required versions.
External MCP
- ✅ Dedicated projects and sessions — External clients can register working directories and create their own sessions. File, CodeKit, shell, and knowledge tools use the workspace bound to the session ID.
- ✅ Background commands and terminals — With the relevant grants, clients can run background commands, inspect or stop jobs, and use terminal sessions from an MCP session.
- ✅ Optional call history — Enable tool call logging per client to review arguments, results, and status. Filter the history by client, project, or session.
Upgrading
- ✅ Standalone MCP calls now require a session ID — Register a project with
vc_mcp_project, create a session withvc_mcp_session, and pass itssidto standalone tools. The session determines the working directory.
Skills and developer verification
- ✅ Managed ROOT skills update at startup — CodeKit, Vulcan Curl, and Vulcan Work update from their configured sources when their installed versions are below the required minimum. A failed update keeps the usable version and is retried at the next start.
- ✅ VMM policy HTTP acceptance check — Developers can run
make test-vmm-httpto exercise the WebUI policy controls together with a local service over HTTP.
Vulcan Code 0.1.27
Vulcan Code 0.1.27 adds a request context inspector, makes skills and plugins easier to invoke on demand, and improves subagent coordination and MCP client grants.
Conversations and workspace
- ✅ Inspect request context — Open Request Context from the conversation menu to review the prompts, messages, and tool definitions used in the latest request. Before the conversation sends its first request, the inspector previews the context it would assemble.
- ✅ Keep the main task moving — The parent conversation can handle priority messages and receive results while subagents are still running. Returned results are included in subsequent requests.
- ✅ Scroll position and elapsed time — Streaming updates preserve your reading position when you scroll back. Reopening a conversation restores the elapsed time of an active request from Core events.
Skills, plugins, and MCP
- ✅ Mention skills and plugins in the composer — Type
@to select a skill or project path, and#to select a plugin. Suggestions show short names. - ✅ Load tools on demand — Vulcan Browser, Context7, and session MCP connection management can be invoked on demand. LuaSkills 0.9.1 lets you switch package tools with guides to on-demand invocation.
- ✅ Grant MCP tools by group — Configure session, standalone, and multimodal permissions separately for each client. Grant access to individual file, shell, CodeKit, and knowledge base tools and set project or library scopes.
Installation, models, and feedback
- ✅ Standalone installer 0.2.18 — On first install, the installer measures response times and recommends the faster download route. You can change the selection, and LuaSkills uses the Hub and GitHub download source for the selected region.
- ✅ LuaSkills startup diagnostics — Startup shows localized progress for dependency setup and identifies the dependency that failed to install.
- ✅ Qwen4 Max — The generally available Qwen4 Max is now listed for Alibaba Cloud Token Plan accounts.
- ✅ Optional feedback contact details — Choose whether to include and preview contact details when submitting feedback. Public replies are available in the in-app message center.
Vulcan Code 0.1.26
Vulcan Code 0.1.26 lets you reconnect a missing project folder without losing sessions, and improves app updates and tool results.
Projects and sessions
- ✅ Recover a project folder — If a folder moves or a drive disconnects, recheck the original location or choose a new folder and keep your session history. The change appears in the conversation timeline, and the next model request uses the new folder.
- ✅ Browse mounted drives and typed paths — The folder picker lists mounted volumes on Linux and macOS. After typing a path, open it before confirming the folder selection.
- ✅ Open the latest message when switching conversations — Switching conversations takes you to the latest message so you can pick up where the conversation left off.
Installation and updates
- ✅ Prepare updates before restarting — Automatic downloads are on by default. The app lets you know when a downloaded and verified update is ready, then switches versions when you confirm the restart. You can still turn off automatic downloads and update manually.
- ✅ Standalone installer 0.2.17 — Windows has a signed graphical installer, Linux has AppImage packages, and macOS has a DMG. After a terminal install on macOS,
vcode manager guiopens the graphical manager. On first install, LuaSkills selects the Mainland China or international Hub that matches the install route.
Browser and document tools
- ✅ Browser action cards — Playwright cards show the action, target, and result for navigation, clicks, screenshots, and other operations. The bundled Vulcan Browser skill guides the assistant through browser actions and verification.
- ✅ Vulcan Work document tools — Vulcan Code installs Vulcan Work from the selected LuaSkills Hub. Its tools work with PDF, Word, Excel, and PowerPoint. Dedicated cards show files, spreadsheet ranges, or results as relevant. Help for implicitly registered LuaSkills tools now lists the parameters each tool accepts.
Vulcan Code 0.1.25
Vulcan Code 0.1.25 improves desktop startup, knowledge bases, and dedicated assistants. The standalone setup program 0.2.13 adds a choice of download routes.
Startup and runtime
- ✅ Desktop startup progress — The startup screen shows the current Core stage and names the skill being downloaded.
- ✅ LuaSkills compatibility — Core uses the bundled Lua runtime and module paths without interference from inherited Lua environment variables. Lua modules can load from Windows installation paths with non-ASCII characters.
Knowledge bases
- ✅ Reuse indexes — Switching search modes preserves existing indexes and embedding settings. Re-enabling the same vector space fills in only the missing content.
- ✅ Document previews and search — The web app previews current and historical documents in Markdown, with an option to view the source. Search summaries open at the beginning of the document. Text matches still open at the matching line.
Assistants and models
- ✅ Locked memory — Converting a session opens the floating assistant and shows its lock status. Turns completed under the lock are marked so you can see which content will be left out of the next turn.
- ✅ System prompts — Changes to the base system prompt take effect on the next execution, before conversation history is sent, without refreshing the session.
Standalone setup 0.2.13
- ✅ Download routes — Choose Automatic, International, or Mainland China when installing or updating. Setup remembers the choice, and you can set it with
--route.
Vulcan Code 0.1.24
Vulcan Code 0.1.24 improves attachments and conversation recovery when switching models, with fixes for the built-in browser and terminal logs.
Conversations and input
- ✅ Drop and paste files — Attachment previews are available again. Files the model cannot upload can be sent as references when the client can read their original paths. Sent file references expand separately from the message body.
- ✅ Model switching and recovery — Switching models preserves the original conversation history. If reasoning cannot be replayed safely, copy the visible messages into a new conversation draft and review it before sending.
- ✅ Compaction and scrolling — Context compaction shows separately while it runs, and new text stays visible when following the bottom of a conversation.
Built-in browser
- ✅ Page rendering and editing — Fixed stylesheet and popup resource loading and improved editing. Passwords stay masked in the built-in browser when the page toggles password visibility.
Logs and providers
- ✅ Terminal logs — Improved redaction of stored secret values across line breaks and clipped output. Background jobs can still be stopped when redaction is unavailable.
- ✅ Provider errors — Error cards retain readable provider messages and help links while masking URLs that contain credentials.
- ✅ Provider compatibility — Alibaba Cloud Model Studio Token Plan no longer attempts temporary OSS uploads. Fixed knowledge base requests that use OpenVulcan embeddings.
Vulcan Code 0.1.23
Vulcan Code 0.1.23 adds protected conversation history cleanup, updates the built-in browser, and fixes bundled Lua modules on Linux and macOS.
Updating to this version
- ✅ Installer 0.2.12 — The install script reuses the existing setup program, which updates and restarts automatically. On Linux and macOS, choosing PATH integration makes the commands available in new terminals. Command-line setup now shows service commands and the WebUI ports.
Conversations
- ✅ History cleanup — Use
/clean-historyor the composer’s add menu in an idle conversation after context compaction. Choose the last turn to clean. Cleanup removes old tool and reasoning details, keeps simplified questions and answers in the timeline, and protects the history the conversation still needs. - ✅ Long conversations — Streaming Markdown updates independently of the rest of the transcript. The timeline preserves your reading position as older entries load, including cleaned history.
Built-in browser
- ✅ Page rendering — WebUI renders the built-in browser’s page structure in place of the previous screenshot stream.
- ✅ Viewport sizes — Choose a desktop, portrait, or landscape preset, enter a custom size, or fit the viewport to the current window with one action.
Workspace and runtime
- ✅ Knowledge base — New sidebar layouts pin Knowledge by default. Your saved layout takes precedence.
- ✅ macOS Lua runtime — Fixed loading of bundled Lua native modules on Intel and Apple silicon. Context7 now initializes its local storage when you first open its settings.
- ✅ Linux Lua runtime — Bundled Lua modules such as OpenSSL now load on ARM64 and x64. Release packaging checks that the files support glibc 2.35.
Vulcan Code 0.1.22
Vulcan Code 0.1.22 adds free cloud memory on supported OpenVulcan sites and improves how conversations handle images, provider failures, and voice recordings.
Memory
- ✅ Free cloud memory — Sign in to OpenVulcan on a supported site. Review the models, processing regions, and data handling terms before enabling it. Reported token usage is broken down by memory stage.
- ✅ Vector rebuild — See completed batches and the retry countdown when rate-limited. Rebuilding resumes after the wait.
- ✅ Markdown memory — Revision conflicts now distinguish changes staged within the batch from the last saved revision.
Conversations
- ✅ New images in long conversations — With automatic compaction enabled, earlier images can be summarized in batches to make room. The new images are then sent in the original request.
- ✅ Provider errors — Bailian quota errors are recognized as rate limits. Balance and network failures appear in error cards, and routine system notices appear as status rows.
Voice input and tools
- ✅ Recording duration — Recordings that reach the time limit can be submitted for transcription.
- ✅ RTK statistics — Command rankings load without scanning Claude conversation logs. They still use the same RTK statistics database.
Vulcan Code 0.1.21
Vulcan Code 0.1.21 improves push-to-talk shortcuts and makes silent recordings easier to handle.
Voice input
- ✅ Push-to-talk shortcuts — Bind a left or right modifier by itself, such as Right Ctrl, or use a punctuation chord such as Ctrl+/. Workbench shortcuts such as Ctrl+M remain available for their assigned actions.
- ✅ Recordings without speech — Silence and empty transcripts show a brief notice. The draft stays as it was.
- ✅ Notifications — Conversation notifications appear above the composer and leave keyboard focus in the input.
Vulcan Code 0.1.20
Vulcan Code 0.1.20 adds OpenVulcan cloud transcription and project-attached folders, and improves queued turn recovery and provider troubleshooting.
Voice input and workspace
- ✅ Cloud transcription — Sign in to OpenVulcan on a site that offers the service. Microphone settings let you choose the site-recommended model, a specific cloud model, or a configured provider’s speech recognition model.
- ✅ Recording shortcut — Set your own voice input shortcut and choose the recognition language and any options supported by the model.
- ✅ Project-attached folders — Add local folders from the project home to browse them in the file panel and open their files from conversations. Attached Git repositories appear in the project’s changes list.
Conversations and memory
- ✅ Queued turn recovery — Saved queued turns resume with their original submission identity and settings. Goals in existing conversations also start only once.
- ✅ Conversation branches — Fork from a completed turn while the conversation is still running. Tool cards retained in the conversation show their final state when execution ends.
- ✅ Memory vector rebuild — Rebuild batches respect the embedding model’s per-request input limit. The preview shows the same batch count used during the rebuild.
Models and troubleshooting
- ✅ Provider model discovery — Fetch candidate chat models from the selected built-in provider account and use your selection to prefill the new model form.
- ✅ Alibaba Cloud models — Token Plan adds Qwen4.0 Max Preview. The Beijing and Singapore Model Studio endpoints add Qwen3.8 models.
- ✅ Error details and retries — Error cards show a short diagnosis and let you copy the full JSON or add the current error to the retry list. Edit individual rules using keywords or regular expressions.
Vulcan Code 0.1.19
Vulcan Code 0.1.19 adds Xiaomi MiMo, updates Alibaba Cloud Token Plan, and makes settings, shortcuts, and usage costs easier to read.
Models and providers
- ✅ Xiaomi MiMo — Sign in through a browser or use an API key. View API account balances and personal plan credits, and use MiMo for chat, speech recognition, speech synthesis, voice design, and voice cloning.
- ✅ Alibaba Cloud Token Plan — The model list and connections for speech recognition, speech synthesis, and video generation have been updated.
- ✅ Model capabilities — Updated input capabilities and context information for selected models on Open Design and OpenCode Go.
Interface and settings
- ✅ Interactive previews — Try appearance, reasoning popups, tool collapsing, session naming, and message sending in Settings.
- ✅ Composer shortcuts — The composer shows what Enter and Ctrl+Enter will do based on the current task and your sending preference. The hint stays visible while you type during a task.
- ✅ Settings navigation — Software Update has separate tabs for updates and release notes. AI Service for Other Tools explains how external tools can use Vulcan Code over MCP.
Usage statistics
- ✅ Cost details — Keep known subtotals when usage data is incomplete, see which service handled each request, and display costs to three decimal places. MiMo and Alibaba Cloud reference prices have been updated.