wise-tui

Wise is a desktop development workbench designed for AI agents, built with Tauri 2, Rust, React 19, TypeScript, and Vite. It uses local Git repositories as the working context, connecting AI chat, terminal, file editing, branch and diff management, worktrees, workflows, and scheduled automation in a single interface. The goal is to extend AI development tasks from conversation into runnable, observable, and reviewable processes. Core capabilities include: 1. Workspace management: Supports quick start with a single repository, as well as organizing multiple repositories into advanced workspaces; each repository can have its own sessions, execution environments, quick actions, requirements, scheduled tasks, and runtime status. 2. Multiple execution environments: Can integrate Claude Code, Codex RPC, Cursor Agent, DeepSeek Harness, Gemini CLI, OpenCode, and Qoder CLI within the same workspace. The README notes that actual availability depends on local CLI installation and account configuration. 3. Sessions and execution: Supports creating, resuming, stopping, and running multiple AI sessions in parallel, with multi-pane layouts, rich text input, file and image context, voice input, quick commands, permission prompts, sandboxing, model configuration, usage viewing, and process management. 4. HUD overlay: A pinned capsule input bar allows switching repositories, sessions, execution environments, and models, submitting or stopping tasks, and viewing runtime status, Git statistics, and completion notifications; global hotkeys can summon it when the app is not focused. 5. Workflows: Use a canvas to organize stages, nodes, dispatch relationships, and execution order, with support for templates, parallel tasks, execution snapshots, acceptance results, failure retries, and run history review. 6. Capability Hub: Manage assistant templates, agent registries, in-repo .agents, Skills, MCP servers, hooks, plugins, and extensions. Extensions can contribute skills, themes, MCP, assistants, settings pages, and lifecycle scripts, and declare permissions. 7. Automation and Channel: Supports configuring Cron tasks, missions, session continuation, and background execution per repository; can perform code reviews on uncommitted changes or relative to the main branch, and receive notifications, receipts, and remote commands via DingTalk, Feishu, WeCom, Telegram, or generic WebSocket. 8. Artifact and browser monitoring: The artifact inspection panel can browse and preview Markdown, diffs, images, PDFs, Office, HTML, and code text. Chrome page monitoring can collect JavaScript exceptions, console messages, API and network failures, Web Vitals, long tasks, slow requests, white screens, crashes, operation traces, SourceMap locations, and screenshots via CDP or the wise-page-monitor extension, and dispatch issues to agents for tracking and fixing. 9. Canvas: Can open HTML, SVG, Markdown, and other content from the artifact center or repository files, previewing and scaling at desktop, tablet, and phone widths; full HTML/SVG code blocks in chat replies can also be switched to an interactive canvas. Scripts run in an isolated sandbox with limits on resource size and quantity. Data is stored locally by default in ~/.wise/, including SQLite databases, repository registration info, window and session state, images, and monitoring evidence. Repository-level extensions can be placed in .wise/extensions/ within the repo, while global extensions are in ~/.wise/extensions/. The README reminds that an agent's file read/write, command execution, and network capabilities are determined by the corresponding CLI, sandbox, Wise permissions, and extension permissions, and API keys need proper management. Local development requires Bun, Rust stable, and Tauri 2 prerequisites. Common commands include bun install to install dependencies, bun run tauri:dev to start desktop development mode, bun test to run frontend tests, and bun run build with bun run tauri:build to produce installation packages, with build outputs in src-tauri/target/release/bundle/. The project is open-sourced under the Apache License 2.0.