v1.0.4

OpenMind Studio v1.0.4

OpenMind Studio v1.0.4 introduces the OpenMindAI coding assistant, a redesigned editor context menu, more built-in developer tools, network deployment improvements, and important stability and security fixes.

New

OpenMindAI coding assistant

• Integrated coding workspace. A dedicated activity-bar panel can read, search, and edit project files; execute commands; work with Git; and fetch web pages. File changes, deletions, and command execution are subject to permission controls. Ask is the default mode, with Accept Edits and Plan modes also available.
• Local inference by default. The assistant uses a locally downloaded OpenMindAI model served through the bundled llama.cpp engine. Local model inference does not require OpenMindAI Cloud. Optional features such as downloads and web access can still make network requests when used.
• Optional OpenMindAI Cloud. Switch between Local and Cloud in Agent Settings. Cloud supports OpenMindAI Super, Ultra, and Lite using a user-provided Developer API key. On Windows, the key is protected with DPAPI. Settings retain the last model used for each mode, display the configurable Cloud endpoint, and provide an account sign-up link.
• Guided setup. A branded setup flow includes model download details, estimated size and memory requirements, and first-run model preloading. With no folder open, the assistant starts from Documents without creating an unnecessary project folder.
• Task queue and session tools. Queue multiple requests for sequential execution, with per-task permissions and execution limits, interruption recovery, a running task list, and Undo for the previous assistant turn.
• Persistent conversations. Chat history and attachments are saved. Optional media and search tools, GPU settings, and a / command center support repeatable workflows.

Expanded editor context menu

The editor now has a comprehensive right-click menu, including historical and comparison views:

• Go to Definition, Declaration, and References, with Peek actions.
• Rename, Refactor, Format, and source-related actions.
• Git comparisons, file history, blame, opening files on their remote, and copying links, patches, or permalinks.
• OpenMindAI actions when the assistant is available: Explain, Review, Add to Chat, inline chat, and dictation.

More languages and offline formatting

• PascalABC.NET: Install through Runtime Manager, run .pas files, and select PascalABC.NET or Free Pascal per project.
• Assembly: Bundled NASM, GCC, and GDB tooling, asm-lsp support, execution of .asm and .s files, and GDB step-through debugging.
• Offline formatters: Expanded bundled formatting with Prettier and plugins for PHP, shell scripts, Dockerfile, XML, SQL, TOML, Java, Svelte, Astro, and Vue; Ruff for Python; gofmt for Go; clang-format for C/C++/Objective-C/C#; and StyLua for Lua. Existing JSON and language-server formatting remain available.

Built-in browser preview

Running HTML or JavaScript can now open an embedded browser next to the editor, including a console. The preview handles stylesheets, relative scripts, local-file fetch requests, SCSS/Sass compilation, and reloads when served files are saved.

Network deployment and shared installations

• Network Deployment: Access from Help > Network Deployment or OpenMindStudio.exe --deploy ... to discover available installations on the local network, run a trusted shared copy, or install from shared setup/portable packages. The transfer workflow includes package size and SHA-256 integrity checks and signature-verification support; official release signing is not yet configured.
• Explicit deployment trust: A computer must trust a deployment before using it. Shared installations are trusted for their specific network path.
• IT validation: A 20-point Network Deployment Test can be exported as a report.
• Shared shortcuts: A shared installation marked Recommended for All Users can create a desktop shortcut once for each Windows user who launches it.
• Per-computer portable profiles: A portable installation shared across PCs can maintain separate databases, settings, and logs for each computer.
• Safer Windows upgrades: The installer detects existing installations, offers in-place Update/Repair, and prevents downgrades.
• Update interface prepared: GitHub Releases checking, status-bar notifications, and a required-update overlay have been implemented and tested, but signed automatic updates are not active in this release.

Collaboration — early preview

A local-first Room Code collaborative workspace (openmind-collab) is included for shared editing over a network. This feature is experimental until physical multi-computer and classroom testing is completed.

Updated

• More accurate performance reporting: The assistant reports GPU offload and CPU fallback based on the actual llama.cpp runtime behavior, rather than assumptions.
• Reasoning and context controls: Reasoning tokens have a separate limit from response tokens, and the engine uses the context size selected for the model.
• Faster, smoother assistant responses: Independent project-file reads can execute in parallel. Stream updates are batched to reduce interface churn and background writes.
• Release packaging: Windows and Linux release layouts, architecture-specific launchers, legal documentation, and installer licence/privacy presentation have been reorganized. macOS build tooling is prepared, although macOS binaries have not been produced.
• Code-signing preparation: A Windows Authenticode signing workflow and docs/code-signing.md are available. A trusted signing certificate is still required before binaries can be signed.

Fixed

Assistant and editor

• Fixed a local inference session that could remain stuck after a cancelled or aborted response stream. The engine now recovers and releases the active turn.
• Fixed Windows path identity handling so short (8.3) and long forms of the same path are not treated as different files.

Collaboration

• Fixed network room loading from a potentially mixed or partial state; joining now uses one consistent database snapshot.

Network installations and shared runtimes

• Fixed failures when one computer requested a runtime while another computer was installing it into the same shared directory; the second computer waits and reuses the completed installation.
• Prevented simultaneous Language Tools installations from overwriting each other's records using cross-computer locking.
• Prevented an already complete shared runtime from being unpacked or deleted when launcher regeneration failed. The runtime is reused if possible; otherwise the operation stops without modifying it.
• Moved Composer cache/home, notebook runner data, and extension downloads to per-computer storage instead of writing into a shared network location.
• Fixed occasional empty-profile startup from a mapped network drive.
• Corrected shared-install desktop shortcuts to use the UNC (\\server\share) path instead of retaining the installing user's mapped drive letter.
• Stopped installer updates from recreating desktop shortcuts that users had deliberately removed.

Linux packaging

• Fixed a broken-pipe error in the Linux release script when reading the build system's glibc version, which could incorrectly report failure after a successful build.

Security and reliability

• Deployment trust is explicit: Package integrity or a valid release signature does not establish whether a particular network host is trusted. Each computer must trust the intended deployment, and shared-install trust is scoped to the exact network path.
• Safer network discovery: Invalid announcements are ignored when an advertised deployment identity does not match its share path or attempts to change a previously discovered deployment type.
• Hardened package serving: Slow request senders are timed out, and files substituted with symbolic links are not served.
• Safer installation: Package files are protected against substitution between verification and installation; older packages cannot downgrade an existing installation.

Installation and upgrade

• Windows and Linux: Install the appropriate v1.0.4 package manually after it is published. Do not rely on an automatic upgrade until a signed update feed is configured.
• Existing installations: On Windows, the installer can detect an existing installation and offer an in-place Update/Repair without a parallel installation.
• Shared network installations: Ask all users to close OpenMind Studio, then run the updated setup on one computer and select the same shared installation folder.
• Older versions: Users of v1.0.3, v1.0.2, or earlier should plan for a manual installation unless a signed update path is explicitly made available.