MewuAI — Personal Agent & AI Creative Canvas

An open-source, LLM-powered desktop app for conversation, visual annotation, creative tools, and desktop actions.

Windows installer  ·  Windows portable  ·  macOS ARM64  ·  Linux x64

简体中文 · English · Website · Release notes · Report an issue

MewuAI is an open-source desktop Agent app for Windows, macOS, and Linux. Select screen content and express what you need through annotations, freehand drawings, and files. Chat, organize references, generate interactive tools, and use plugins to act in the same space. Mewu with you — working alongside you.

_{Capture, ask, and annotate in place.}

• Personal Agent: choose a role, model, tools, and memory. Reference images, videos, and files; retain conversation history; read Markdown, formulas, and collapsible public reasoning. Use the official Agent's persistent local memory or connect an independent memory service.
• AI creative canvas: arrange, resize, pin, or save images, videos, text, and generated objects. Describe an HTML / SVG / Canvas tool, use the resulting calculator, experiment, or game, and ask for further changes.
• Capture and multiple selections: capture a region, multiple areas, scrolling content, or supported windows. Selections automatically become references while the surrounding screen dims, with no separate “Reference” click.
• Drawing and annotation: pens, arrows, shapes, text, numbering, highlights, and mosaic, with undo, redo, and object editing. Add marks yourself or ask AI to explain and highlight specific details in worksheets, code, and other references.
• Text and tables: local OCR, capture translation, chart understanding, and table extraction. Windows/macOS also offer actionable recognition of links, emails, phone numbers, QR codes, and barcodes.
• Recording and video annotation: capture system audio and microphone input; pause, resume, trim, and export MP4 / MP3 / GIF. Add rectangles and text for time ranges, then click conversation timestamps to revisit them. Audio support varies by system.
• Blackboard: draw and explain on a full-screen board, combining questions, formulas, and AI analysis. A setting lets screen-sharing apps see selections and annotations.
• CLI and region AI operations: local Agents can inspect scenes, capture, add selections, draw, and export. The region plugin lets a model observe a selected target, act, and check a fresh screenshot before continuing.
• Plugins and event triggers: extend Agents, memory, tools, and UI independently. Official Hooks supports schedules, prompts, completed replies, global shortcuts, and app/plugin events, connecting Mewu operations, MCP, commands, and keyboard/mouse actions.
• Skills and self-made plugins: the built-in Agent can read portable SKILL.md skills, create and validate plugin projects, and prepare an installable package. Review permissions before installation under Installed → Self-made. Plugins can reorder the original screenshot toolbar through the generic layout service. See skills and authoring.

The native Agent loads image/video annotation skills on demand, keeps original conversation history when compacting context, and exposes paginated session recall separately from long-term memory. See Agent runtime and repeatable evaluation.

Capture, drawing, and recording work without AI setup. AI conversations, translation, and generated content require a configured service.

• Reference-guided generation and dedicated image/video model connections.
• Circle-to-search, a desktop companion plugin, and more desktop collaboration workflows.
• Broader platform coverage, language support, and performance improvements.

Exam annotation Explain a worksheet and add construction lines, reasoning, and annotations to the original.	Generate HTML tools Describe a tool, generate interactive content, and keep using and refining it.
AI creative canvas Arrange storyboards, images, and conversations to develop a creative project.	Blackboard Draw axes, curves, and notes by hand to explain your reasoning.
Video annotations Annotate time ranges, play the result, and click timestamps to revisit the relevant frames.	CLI capture and drawing Call capture, selection, and drawing operations from a terminal and see the results in Mewu.
Text recognition Extract the text you need from a capture using local OCR.	Code explanation Select code, explain its logic, mark the failing line, and show a correction beside the original.
Table extraction Turn a captured table into structured data you can copy, compare, and use for further analysis.	External Agents Connect Codex, MiniMax Code, WorkBuddy / CodeBuddy, or Hermes to the existing conversation bar.

1. Install and launch. Choose your platform above. Windows uses WebView2; on macOS, move the app into Applications; on Linux, make the AppImage executable. Downloads follow the latest stable release.
2. Enter the workspace. Open Mewu from the tray or a capture shortcut to select, draw, save, or open a blackboard. Configure capture shortcuts in Hooks and grant system capture/recording permissions when requested.
3. Connect AI. Add a model in Settings → Agents & Connections, choose a role and memory, or connect a client through External Agents. Selections become references automatically; send your task from the conversation bar.

For Linux dependencies and permissions, see the AppImage guide and recording guide. Region AI operations support Windows, macOS, and compatible GNOME 50 Wayland x64 setups; Linux also requires Desktop Witness and Portal authorization.

Image, video, and tool capabilities, limits, and fees depend on the selected service. External Agents require the corresponding client to be installed and signed in; consult each plugin's documentation for platform support.

Browse, install, and manage extensions in Settings → Plugins, or import a release ZIP. Browse the full marketplace.

Start the desktop app, then open a terminal. The CLI shares the same Mewu workspace with you:

mewu                         # Start an Agent conversation
mewu status                  # Inspect the app and current scene
mewu schema                  # Discover operations and parameters
mewu call capture.open --json '{}' --wait

Plain mewu chat sends text by default; screenshots and files require explicit references. See the CLI guide and operation catalog for setup, platform requirements, and parameters.

Bug reports, use cases, code, plugins, and documentation contributions are welcome. The desktop app uses Rust + Tauri 2 + SolidJS / TypeScript + SQLite and renders through the system WebView.

Desktop development · Extension SDK · Extension interfaces · Plugin packaging · Release guide

Browser previews support interface development; native capture, recording, and system input require the desktop runtime.

MewuAI's own source uses MPL-2.0, which permits commercial use subject to its terms. Distributions must provide covered source and preserve copyright and license notices as required. Third-party notices ship with the app.

Thanks to Lin Wenjie (林文杰) for subject-matter guidance, Zhang Wenbiao (张文彪) for feature-development suggestions, and blueling0 for testing.