Skip to content

Releases

OpenHub AI v0.2.0

Choose a tag to compare

@github-actions github-actions released this 11 Oct 07:24
852742a

Maintenance and safety notice (October 2026): v0.2.0 remains the latest public release, but new releases are paused while known safety and data-preservation issues are reviewed. Fixes on main are not included in this download. Before installing or updating, read the v0.2.0 known issues. If you use Memory MCP, back up memory.jsonl before changing versions or clearing the npm cache (#69). OpenHub starts third-party MCP code with your user permissions; approval and environment filtering are not an OS sandbox. Avoid sensitive projects or credentials until you have assessed these limits. Report undisclosed vulnerabilities privately. This notice does not change the v0.2.0 binaries or claim that v0.2.1 has been released.

유지보수·안전 안내(2026년 10월): v0.2.0이 최신 공개 릴리스이며, 알려진 안전·데이터 보존 문제를 검토하는 동안 새 릴리스는 보류합니다. main의 수정은 이 배포 파일에 포함되지 않습니다. 설치·업데이트 전에 v0.2.0 알려진 문제를 확인하세요. Memory MCP를 사용한다면 버전 변경이나 npm cache 정리 전에 memory.jsonl을 백업하세요(#69). OpenHub는 승인 후 외부 MCP 코드를 사용자 권한으로 실행하며 OS sandbox가 아닙니다. 민감한 프로젝트·자격증명에 사용하기 전에 한계를 평가하고, 미공개 취약점은 비공개 신고해 주세요.

OpenHub AI v0.2.0

OpenHub AI v0.2.0 makes project analysis cover more real project shapes, adds an English desktop interface, and gives you more control over where and for which client a tool is installed. The Verified Registry has 9 tools. Nothing is installed or changed without a plan you approve, as before.

Project analysis and recommendations

  • New technologies are recognized from project files: Go, Express, NestJS, Jest, Vitest, pytest, Playwright, Unity, Unreal Engine and Kubernetes. Detection uses manifests, dependencies and engine or tool configuration files only; there are no network calls during analysis.
  • Two capabilities were added: Kubernetes operations and game engine editor integration. No game engine editor tool is in the Verified Registry yet, so Unity and Unreal Engine projects show this need without a candidate and say why.
  • When there are no recommendations, the app explains the reason (for example no recognized stack, no verified tool for the need, or no tool that supports your clients or platform) instead of showing an empty list.
  • Recommendations explain why each tool was suggested, and a tool you already use can be added to another client or scope from the same view.

Verified Registry: 9 tools

v0.2.0 ships with 9 verified tools: Playwright MCP, Chrome DevTools MCP, GitHub MCP Server, Context7, Serena, Postgres MCP and Memory from v0.1.1, plus two new ones:

  • MongoDB MCP Server, configured read-only with telemetry disabled.
  • Kubernetes MCP Server, limited to the reviewed version 0.0.67, read-only, with the core toolset and an OpenHub-generated tool configuration that denies reading Secrets. Use it with a least-privilege account: Pod and Node logs it can read may still contain sensitive data.

Every tool passed the same verification criteria as before. The earlier target of 25–40 verified tools is not met in this release and is carried over to a later version with the same criteria; unverified tools are not added to reach a number.

Installation and lifecycle

  • Client choice: choose which clients (Claude Code, Codex, Cursor) a tool is installed for. Each client shows whether OpenHub has verified that it starts the tool on your operating system.
  • Project or user scope: install for the current project (default) or for your user account (Codex and Cursor). A user-scope install affects every project that uses that client and needs its own approval; the plan shows the exact files that will change.
  • Installed tools by scope: the Installed view shows project and user installs separately, with status, Health checks and repair for each.
  • Repair: a missing or changed OpenHub-generated tool configuration, or an invalid client launcher on Windows (for example after Node.js was moved), is detected and can be repaired after you approve a repair plan. OpenHub does not overwrite changes made by others in the meantime.
  • npx packages are prepared before configuration: exact-version npx packages are downloaded and checked after approval and before any client file is written, so a slow first download no longer makes the Health check time out.
  • Installs are decided per target (tool, client, scope and server name), so adding a client or scope later does not touch the existing ones.
  • Update to an exact version (npx tools only): in the Installed view, a tool installed with npx has an optional target version. Enter an exact version in the form X.Y.Z (for example 1.2.3); ranges, tags such as latest, pre-release versions, URLs and paths are rejected before any plan is made. Leave it empty to keep the previous behavior (the Registry's version). The update still needs your approval in the confirmation dialog, which shows the current and target version; the package is prepared before any client file changes, and if the Health check fails, the client configuration and Version State are restored. Kubernetes MCP Server stays limited to its reviewed version 0.0.67.

Desktop language

  • The desktop app is available in English and Korean. On first start it follows your operating system language (Korean for Korean, English otherwise); a language you choose is saved and kept across restarts.
  • Approval dialogs and security notices are shown in the selected language. A notice that has no translation yet is shown in its original text and marked as untranslated, never hidden.
  • The CLI output is unchanged in this release.

Compatibility

  • No breaking changes to CLI commands, client configuration files, the Registry Manifest format, or the install and lifecycle plan formats. Plans, Version State and settings from v0.1.x remain valid.
  • Recommendation reports (openhub project recommend --json) now report taxonomyVersion: 2; the report schemaVersion is still 1. Saved reports from v0.1.x should be regenerated.
  • Desktop users without a saved language choice may now see English on first start if their operating system language is not Korean. The language can be changed at any time.

Known limitations

  • Installing is not the same as a verified start. Per-client start verification exists for Kubernetes MCP Server only; the other 8 tools have no per-client run record. The install plan shows the level for each client and operating system.

    Kubernetes MCP Server, started by the real client from the configuration OpenHub wrote:

    Client Windows Linux macOS
    Claude Code verified verified not verified (platform unverified)
    Codex verified verified not verified (platform unverified)
    Cursor not verified not verified not verified (platform unverified)
  • Cursor: OpenHub writes Cursor configuration, but starting tools from Cursor is not verified on any platform.

  • Claude Code: project scope only; OpenHub does not write the Claude Code user configuration.

  • macOS: no official artifact and nothing is verified on macOS (platform-unverified). Building from source is not blocked.

  • Kubernetes MCP Server: only version 0.0.67 is allowed. Other versions are blocked until reviewed.

  • Windows installer: the NSIS installer has no code signature. How Windows SmartScreen or Smart App Control reacts to it was not observed in release testing; a warning may appear.

  • Linux AppImage: release testing ran the AppImage on CI runners without FUSE by extracting it (--appimage-extract) and starting the extracted app. Starting the AppImage file directly through FUSE was not part of release testing.

Downloads

Windows x64

Download:

OpenHub-AI-Setup-0.2.0-x64.exe

The Windows installer is currently unsigned. Windows SmartScreen or Smart App Control may display a warning.

Linux x64

Download:

OpenHub-AI-0.2.0-x86_64.AppImage

CLI

Download:

openhub-ai-0.2.0.tgz

Install it with:

npm install -g ./openhub-ai-0.2.0.tgz
openhub --version
openhub registry list
openhub project scan ./my-project

Node.js 24.15 or later is required.

Security and Verification

This release includes:

  • SHA256SUMS for release artifact integrity
  • CycloneDX 1.7 dependency SBOMs
  • Windows native/runtime artifact SBOM
  • Electron runtime provenance verification
  • License and third-party notice validation
  • CLI, Windows, and Linux smoke tests

The packaged desktop applications use Electron 44.5.1.

Supported Platforms

Platform Support
Windows x64 NSIS installer
Linux x64 AppImage
macOS Source build only; no official artifact in v0.2.0

OpenHub AI v0.1.1

Choose a tag to compare

@github-actions github-actions released this 09 Oct 14:57
95ed290

OpenHub AI v0.1.1

OpenHub AI v0.1.1 is a patch release focused on security hardening and release reliability. It adds no new features and contains no breaking changes.

Security

  • Fixed a regular expression denial of service (ReDoS) in URL credential detection. OpenHub detects and redacts scheme://user:password@ credentials in text it stores or shows. The pattern used for this took time quadratic in the input length, and it is applied to third-party text such as npm deprecated messages, npm and PyPI runtime requirements, GitHub release titles and discovery descriptions. In our measurements, a 128 KiB npm deprecated message took about 8 seconds to process when checking releases (openhub release and the Desktop release view), and the time grew quadratically with longer input. Detection now runs in linear time and returns the same results as before.

Release reliability

These changes affect how OpenHub AI releases are built and published. They do not change the application itself.

  • After uploading, the release workflow reads the GitHub Release back through the GitHub API. It requires all eight release assets to be present exactly once, fully uploaded, and of the expected size, with GitHub digests that match SHA256SUMS and the built files.
  • The workflow no longer creates a second Release for the same tag. It stops if a published Release or more than one draft already exists, and it reuses a single existing draft when rerun.
  • Rerunning the workflow on an existing draft re-uploads the assets only. Release notes already written on the draft are kept.
  • A new workflow checks a Release when it is published: required assets, digests against the published SHA256SUMS, and a single published Release for the tag.
  • User-facing release notes for each version are kept in docs/release-notes/ and used when the draft Release is created.

Build and CI

  • All build and release scripts are now type-checked in CI. One existing type error was fixed; script behavior is unchanged.

Compatibility

  • No breaking changes. CLI commands, Desktop behavior, configuration files, the Registry format and data schemas are unchanged from v0.1.0.

Downloads

Windows x64

Download:

OpenHub-AI-Setup-0.1.1-x64.exe

The Windows installer is currently unsigned. Windows SmartScreen or Smart App Control may display a warning.

Linux x64

Download:

OpenHub-AI-0.1.1-x86_64.AppImage

CLI

Download:

openhub-ai-0.1.1.tgz

Install it with:

npm install -g ./openhub-ai-0.1.1.tgz
openhub --version
openhub registry list
openhub project scan ./my-project

Node.js 24.15 or later is required.

Security and Verification

This release includes:

  • SHA256SUMS for release artifact integrity
  • CycloneDX 1.7 dependency SBOMs
  • Windows native/runtime artifact SBOM
  • Electron runtime provenance verification
  • License and third-party notice validation
  • CLI, Windows, and Linux smoke tests

The packaged desktop applications use Electron 44.5.1.

Supported Platforms

Platform Support
Windows x64 NSIS installer
Linux x64 AppImage
macOS Source build only; no official artifact in v0.1.1

OpenHub AI v0.1.0

Choose a tag to compare

@kanghyunsoon kanghyunsoon released this 09 Oct 11:07
2e712ee

OpenHub AI v0.1.0

OpenHub AI is an open-source, project-aware lifecycle manager for AI coding tools.

It analyzes your project, recommends relevant MCP servers and agent tools, and helps you install, verify, update, adopt, benchmark, and roll them back with explicit approval before anything runs.

Highlights

  • Project analysis for languages, frameworks, databases, infrastructure, and AI clients
  • Project-aware recommendations with Project Fit and OpenScore
  • Install support for npx, uvx, and Docker backends
  • Claude Code, Codex, and Cursor configuration support
  • Version State, health checks, update recovery, and rollback
  • Adopt existing MCP configurations without rewriting them
  • Release summaries and update impact analysis
  • Discover views for New, Trending, Verified Registry, and Unverified Candidates
  • MCP startup benchmarking without calling tools
  • CLI and Electron desktop app

Downloads

Windows x64

Download:

OpenHub-AI-Setup-0.1.0-x64.exe

The Windows installer is currently unsigned. Windows SmartScreen or Smart App Control may display a warning.

Linux x64

Download:

OpenHub-AI-0.1.0-x86_64.AppImage

CLI

Download:

openhub-ai-0.1.0.tgz

Install it with:

npm install -g ./openhub-ai-0.1.0.tgz
openhub --version
openhub registry list
openhub project scan ./my-project

Node.js 24.15 or later is required.

Security and Verification

This release includes:

  • SHA256SUMS for release artifact integrity
  • CycloneDX 1.7 dependency SBOMs
  • Windows native/runtime artifact SBOM
  • Electron runtime provenance verification
  • License and third-party notice validation
  • CLI, Windows, and Linux smoke tests

The packaged desktop applications use Electron 44.5.1.

The Windows executable is branded as OpenHub AI, so Syft may not identify the executable itself as Electron. Electron identity is instead verified through the dependency SBOM, the running application, and the official Electron distribution.

Security Hardening Before Release

Before v0.1.0 was published, release-note parsing was changed to a linear-time parser after CodeQL identified two polynomial backtracking paths reachable from third-party release notes.

The fix is included in this release.

Supported Platforms

Platform Support
Windows x64 NSIS installer
Linux x64 AppImage
macOS Source build only; no official artifact in v0.1.0

Important Behavior

OpenHub does not treat recommendations, release notes, scores, AI summaries, or discovered candidates as permission to execute anything.

Install, update, rollback, adopt, and benchmark operations use explicit plans and approval checks before effects are performed.

Unverified Candidates cannot be installed, adopted, updated, or used as actionable recommendations.

Documentation

See the repository README for Quick Start, CLI commands, Desktop usage, architecture, security model, approval model, troubleshooting, and contribution guides.

This is the first public release of OpenHub AI.