A local-first desktop workspace for AI-powered projects, research, documents, and learning.
Run local OpenAI-compatible models, use your existing Grok or Codex CLI session, and keep every conversation and file organized by project.
LocLM is an open-source Electron desktop application that brings local and subscription-backed AI into one focused workspace. Projects combine conversations, reusable files, model settings, web research, screen captures, and learning games without turning the interface into a developer console.
Local models remain the default path. External providers are explicit: Grok uses the existing Grok CLI session, Codex uses the installed Codex CLI and its existing OpenAI login, and web search only runs when enabled for a message.
| Area | Capabilities |
|---|---|
| AI providers | LM Studio and other OpenAI-compatible local endpoints, Grok subscriptions, and Codex CLI |
| Project context | Project-scoped chats and files; extracted document text and supported images are automatically available to the selected AI |
| Chat experience | Streaming responses, cancellation, model switching, Markdown, source cards, and collapsible reasoning/work summaries |
| Documents | Read PDF, DOCX, text, Markdown, JSON, CSV, PNG, JPEG, and WebP; export responses to Word or PDF |
| Research | Built-in background web search with DuckDuckGo/Bing fallback, plus optional Brave Search and SearXNG |
| Screen capture | Global, remappable shortcut with optional automatic analysis and temporary-image handling |
| Learning games | AI-generated quizzes, fill-in-the-blank tasks, matching exercises, exam simulations, local scoring, and detailed answer review |
| Desktop experience | Light, dark, and system themes; English and Hungarian UI; automatic updates through GitHub Releases |
| Learning results | Project file context |
|---|---|
 |
 |
| Codex CLI provider | Web search configuration |
|---|---|
 |
 |
Adding or replacing screenshots and GIFs
Keep repository media in docs/screenshots/ and use descriptive, lowercase filenames. The automated smoke test can regenerate the current screenshots:
$env:LOCLM_SMOKE_SCREENSHOT_DIR = "docs/screenshots"
pnpm test:smokeFor an animated walkthrough, add docs/screenshots/loclm-demo.gif and replace the main screenshot near the top of this README with:
For a clean GitHub preview, keep media at a 16:9 or similar desktop aspect ratio and optimize large GIFs before committing them.
Download LocLM Setup 0.1.7 for Windows, or open the complete v0.1.7 release page, then run the installer.
The current community builds may be unsigned. If Windows SmartScreen appears, verify that the download came from this repository's Releases page before choosing More info → Run anyway.
LocLM does not bundle an AI model. Configure at least one provider after installation:
- A local OpenAI-compatible server such as LM Studio.
- An authenticated Grok CLI session.
- An installed and authenticated Codex CLI.
- Open Settings → AI.
- Choose Local AI, Grok, or Codex.
- Test the connection and select a model.
- Create a project and add any files you want available as reusable context.
- Start a chat, enable web research when needed, or create a learning game from your source material.
Start an OpenAI-compatible server and enter its base URL in LocLM. For LM Studio, the default is usually:
http://127.0.0.1:1234/v1
LocLM lists the available models from the endpoint. Enable vision support for models that can process images and screen captures.
LocLM reuses the Grok CLI session stored on the computer. Sign in with grok login or use the sign-in action in Settings → AI → Grok, then select one of the models exposed by your subscription.
Prompts and attachments sent through Grok are processed by xAI through the authenticated Grok CLI service.
Install Codex CLI and run codex login, or start the ChatGPT sign-in flow from Settings → AI → Codex. LocLM invokes codex exec --json in an ephemeral, read-only session and does not read or copy your Codex credentials.
If the executable is not on PATH, set LOCLM_CODEX_PATH to the full Codex executable path before starting LocLM.
The built-in provider performs API-key-free searches in an isolated background Chromium window and sends only cleaned titles, URLs, and snippets to the selected model. Brave Search and self-hosted SearXNG are also supported.
Web access is message-scoped: turn on Web in the composer when current information is needed. Search results are preserved as source cards alongside the assistant response.
- The renderer has no direct Node.js access; Electron context isolation and sandboxing are enabled.
- The preload bridge exposes a typed, restricted set of IPC operations.
- Local model prompts and attachments stay on the computer unless an explicitly selected external provider or web search is used.
- API keys are stored through Electron's encrypted credential storage when available.
- Codex and Grok authentication is handled by their respective CLI sessions.
- Only HTTP and HTTPS URLs can be opened externally.
- Node.js 24 or newer
- pnpm 11 or newer
- Windows, macOS, or Linux desktop environment
git clone https://github.com/TheRealMagyar/LocLM.git
cd LocLM
pnpm install
pnpm devpnpm typecheck
pnpm build
pnpm test:smoke# Current operating system
pnpm dist
# Windows installer
pnpm build
pnpm exec electron-builder --win --publish neverBuild output is written to release/.
GitHub Actions builds release artifacts from version tags. After updating the package version and validating the application:
git tag -a v0.1.7 -m "LocLM v0.1.7"
git push origin main
git push origin v0.1.7The release workflow publishes the Windows installer and update metadata to GitHub Releases. Production distribution should use code signing through the configured CSC_LINK and CSC_KEY_PASSWORD repository secrets.
src/main/ Electron main process and desktop services
src/preload/ Typed, isolated IPC bridge
src/renderer/ React user interface
src/shared/ Shared TypeScript types and model utilities
scripts/ Automated smoke and integration tests
docs/ README media and project documentation
build/ Application icons and packaging resources
Issues and pull requests are welcome. For UI changes, include before/after screenshots where practical and run the quality checks above before submitting.
LocLM is available under the MIT License.