Cadenza
williammb/cadenza
Quadro de tarefas desktop local para agentes de IA com interface desktop e CLI integrados
Ver no GitHubGerenciador de tarefas desktop local-first construído em Tauri + Rust para colaboração entre humanos e agentes de IA (Claude Code, Codex). Oferece tanto app desktop quanto CLI, com comunicação segura via named pipes/Unix sockets e interface de quadro Kanban para revisão humana de propostas de trabalho.
Instalação
npx convez-ai williammb/cadenza1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Colaboração entre desenvolvedores e agentes IA em projetos Rust/Tauri
- Automação de fluxos de trabalho com revisão humana intermediária
- Gerenciamento de tarefas com histórico e rastreamento persistente
Veredito

Solução robusta e bem arquitetada para coordenar agentes IA com humanos em um loop de feedback. O design local-first e o protocolo NDJSON autenticado garantem segurança; ainda em pré-release, mas código de produção e documentação clara.
README
A tradução para o português deste README ainda está na fila.
Cadenza
A local-first desktop task board for AI coding agents.
Status: public pre-release. The repository is open for development, but stable public binaries are not published yet.
Version: tag-driven — main carries a 0.0.0 placeholder; each release's version comes from its vX.Y.Z git tag (see docs/RELEASING.md)
Stack: Tauri 2, Rust, vanilla HTML/CSS/JS
Release artifacts: Windows NSIS first. Linux AppImage and macOS DMG are planned but not published yet.
License: MIT OR Apache-2.0
What Cadenza Does
Cadenza is a desktop app that helps a human collaborate with AI coding agents through a local task board. It ships two binaries from the same installer:
cadenza: the Tauri desktop app with tray, webview UI, notifications, task storage, terminal sessions, and update checks.cadenza-cli: the command-line client that agents use to read tasks, report progress, propose work, and mark work as done.
The app and CLI communicate on the same machine over an authenticated named pipe on Windows or Unix socket on Linux/macOS. The protocol uses NDJSON request/response messages. Cadenza does not open a TCP server for agent communication.
Agents do not directly create or complete task scope. They submit proposals, and the human accepts or rejects them in the app.
Current Features
- Local task board with the states
a_fazer,fazendo,aguardando_revisao, andfeito. - CLI for AI agents, with stable exit codes for automation.
- Idempotent proposal flow with persisted recovery data.
- PTY-backed agent runs and terminal streaming through Tauri channels.
- Projects and active-project tracking.
- Ideias inbox.
- File, SQLite, and PostgreSQL storage backends.
- PostgreSQL password storage through the OS keyring.
- Shared i18n through Fluent files in
locales/. - Vanilla HTML/CSS/JS UI with vendored browser libraries.
- Windows NSIS packaging path and signed updater plumbing.
Architecture
cadenza desktop app
ui/ static HTML/CSS/JS loaded by Tauri
src-tauri/ Rust backend, tray, commands, store, IPC, PTY
|
| Tauri invoke / emit / Channel
|
webview UI
cadenza-cli
|
| NDJSON over named pipe / Unix socket
|
cadenza IPC server
| Channel | Direction | Purpose |
|---|---|---|
Tauri invoke | Frontend to backend | Task CRUD, projects, settings, decisions |
Tauri emit | Backend to frontend | Real-time UI updates |
Tauri Channel | Backend to frontend | PTY terminal streaming |
| Named pipe / Unix socket | CLI to app | Authenticated NDJSON protocol |
Repository Layout
cadenza/
src-tauri/ Tauri app: tray, IPC server, store, triage, PTY, notify
cadenza-cli/ clap CLI that talks to the app over the local socket
proto/ shared NDJSON wire types
i18n/ shared Fluent bundle loader and locale resolution
locales/ .ftl files for app, CLI, and UI
ui/ static HTML/CSS/JS
ui/vendor/ vendored browser libraries
skills/ per-locale agent skill snippets
skills-core/ shared skill metadata and loader
docs/ release and maintenance docs
installers/ packaging support files
Cargo.toml workspace manifest
deny.toml cargo-deny policy
AGENTS.md Codex guidance for this repository
CLAUDE.md Claude Code guidance for this repository
Requirements
- Rust stable >= 1.77
- Tauri CLI 2:
cargo install tauri-cli --version "^2" --locked
- Platform dependencies required by Tauri 2:
- Windows: WebView2
- Linux: WebKitGTK and AppIndicator/Ayatana development packages
- macOS: Xcode command line tools
See the official Tauri prerequisites: tauri.app/start/prerequisites
The UI has no package-manager workflow and no frontend build step. Browser
libraries are vendored under ui/vendor/ and documented in
THIRD_PARTY_NOTICES.md.
Major Rust dependencies include Tauri 2, Tokio, SQLx, Fluent, Interprocess, portable-pty, keyring, clap, tracing, serde, and cargo-deny policy checks. The authoritative dependency list is in Cargo.toml and Cargo.lock.
Development
Run the desktop app:
cargo tauri dev
Build the current platform bundle:
cargo tauri build
Run the CLI from source:
cargo run -p cadenza-cli -- current --json
Run tests:
cargo test
Run lint and policy checks used by CI:
cargo fmt --check
cargo clippy --workspace --all-targets --all-features -- -D warnings
cargo deny check
CLI Example
cadenza-cli requires the Cadenza app to be running.
cadenza-cli current --json
cadenza-cli list --estado doing
cadenza-cli log T-42 "implemented the handshake"
cadenza-cli propose --parent T-42 \
--title "extract auth module" \
--repro "..." \
--file "src/auth/mod.rs" \
--what-failed "..." \
--action "..."
cadenza-cli done T-42 "merged, tests green"
--estado accepts English aliases (todo, doing, review, done).
JSON output uses the canonical Portuguese state values:
a_fazer, fazendo, aguardando_revisao, and feito.
Exit Codes
| Code | Meaning |
|---|---|
| 0 | ok |
| 1 | generic error |
| 2 | bad usage |
| 10 | app not running |
| 11 | bad or missing token |
| 12 | protocol mismatch |
| 20 | proposal rejected |
| 21 | decision timeout |
| 30 | task not found |
Project Rules
- The UI is plain HTML/CSS/JS served by Tauri. Keep it framework-free and build-step-free.
- Backend, CLI, and UI share Fluent locale files. Do not add a second i18n system.
- The IPC protocol is NDJSON over local OS IPC, with
protocol_versionnegotiated separately fromapp_version. proposemust stay idempotent and resumable through client-generated idempotency keys.- UI code must avoid
innerHTMLexcept inui/markdown.js, where markdown output is sanitized before assignment. - Dependencies must remain compatible with the repository license policy:
MIT, Apache-2.0, BSD, or MPL-2.0.
cargo-denyenforces this in CI.
Agent-specific implementation guidance lives in AGENTS.md and CLAUDE.md.
Privacy and Local Data
- Cadenza stores local app data under the user's Cadenza data directory, including task data, inbox items, proposals, logs, and the local auth token.
- The CLI token is used only for same-host communication between the app and CLI.
- PostgreSQL passwords are stored in the OS keyring when the PostgreSQL backend is used.
- Public release builds may contact the configured GitHub release endpoint to check signed updater metadata.
Contributing
Read CONTRIBUTING.md before opening a PR.
Portuguese is welcome for architecture discussions. English is fine for code, comments, commits, issues, and PRs.
Security issues should be reported privately; see SECURITY.md.
License
Dual-licensed under MIT or Apache-2.0, at your option.