A desktop app that turns vague ambitions into goals, tasks, and evidence — powered by your local coding agents.
Modern coding agents (Codex, Claude Code, Kimi CLI, etc.) are powerful but isolated. Each has its own interface, session history, and workspace. When you're working toward a larger objective — shipping a feature, learning a framework, refactoring a codebase — you end up context-switching between tools and losing track of the bigger picture.
Goalward bridges this gap:
- Goal-driven workflow: Start with a direction, not a command. Goalward helps you clarify outcomes, success criteria, and break work into verifiable tasks.
- Unified agent proxy: Run multiple coding agents from one place. Native protocol support means streamed replies, tool progress, and approvals work as the agent intends.
- Evidence-based review: Every execution is traced. Artifacts, conversations, and results stay linked to their task. You decide what counts as done.
- Your agents, your machine: Goalward manages agent processes locally. No cloud relay — your code and credentials stay where they are.
- Goal Assistant: Describe a broad direction; a restricted local agent helps clarify outcome, scope, and success criteria, then proposes tasks.
- Task Management: Create, edit, and track tasks linked to goals. Set deadlines, acceptance criteria, and dependencies.
- Multi-Runtime Execution: Run Codex, Claude Code, Kimi CLI, Pi, DeepSeek Harness, or custom CLI tools. Switch agents per task or combine them.
- Native Sessions: Conversations persist across turns. Connections are reused; session IDs are preserved for exact resumption.
- Streaming & Approvals: Watch agent replies arrive in real time. Handle tool approvals, file changes, and permission requests through the native UI.
- Execution Traces: Every run is recorded — input, output, tool calls, reasoning summaries, token usage, and timing.
- Artifact Preview: Review HTML, Markdown, images, PDFs, and web pages directly in the inspector panel.
- Themes & Appearance: System, Dark, Light, or Cobalt2 themes. Background images with adjustable opacity. Respects reduced motion and transparency preferences.
- In-App Updates: Check for updates, download signed packages, and install without leaving the app.
| Agent | Protocol | Streaming | Session Continuity | Input While Running |
|---|---|---|---|---|
| Codex | app-server stdio:// |
✓ | ✓ Thread resume | ✓ Steer |
| Claude Code | Agent SDK bidirectional | ✓ | ✓ Session resume | ✓ Follow-up queue |
| Kimi CLI | ACP | ✓ | ✓ Session resume | Draft retained |
| Pi | RPC mode | ✓ | ✓ Session resume | ✓ Steer |
| DeepSeek Harness | CLI | ✓ | — | — |
| Custom CLI | Configurable | Varies | — | — |
Goalward uses the original agent executables. You install and authenticate them yourself; Goalward manages the process lifecycle and protocol integration.
Goal overview with task delivery progress, success criteria, and working directories:
Tasks workspace with conversation, execution inspector, and artifact preview:
Appearance settings with theme, background, and language controls:
Screenshots show the current frontend at 1536 × 960 with sample data. They do not show live agent execution.
Goalward runs on Apple Silicon and Intel Macs with macOS 13.3 or later. Each download is a universal app (arm64 + x86_64).
Every version remains on GitHub Releases.
Automatic updates read https://releases.goalward.dev/latest.json first and fall back to the GitHub Release if that manifest is unavailable. Releases include SHA-256 checksums. The app uses ad-hoc signing and is not Apple-notarized; macOS will prompt on first launch. See the installation guide for details on bypassing Gatekeeper.
-
Download and install. Open the DMG, drag Goalward.app to Applications, and launch it. If macOS blocks the app, go to System Settings → Privacy & Security and click "Open Anyway."
-
Connect an agent. On first launch, Goalward detects installed runtimes. Import a detected agent or configure one manually in Settings → Runtimes. Make sure your agent CLI is installed and authenticated.
-
Create a goal. Click "Start something" and describe what you want to accomplish. The goal assistant helps clarify the outcome and success criteria, then proposes tasks.
-
Execute tasks. Open a task, select an agent and model, and send instructions. Use
@to reference files. Watch the streamed response, handle approvals, and review artifacts. -
Review and continue. Check execution traces and artifacts. Accept the task when criteria are met, or iterate. Mark the goal achieved when all success criteria have evidence.
Prerequisites: macOS 13.3+, Node.js (20.19+, 22.12+, or 24+), Rust stable, Xcode Command Line Tools.
git clone https://github.com/Beace/goalward.git
cd goalward
npm ci --registry=https://registry.npmjs.org
npm run mac:dev # Start development app with hot reloadOther commands:
npm run mac:build— Build the app bundlenpm run build— TypeScript + Vite buildnpm test— Run frontend testscargo test --manifest-path src-tauri/Cargo.toml— Run Rust tests
See CONTRIBUTING.md for the full development setup and PR workflow.
Goalward is designed around a specific workflow: Goal → Tasks → Agent Execution → Evidence → Review.
- Goal-first, not chat-first. You start with an outcome and success criteria, not an open-ended conversation. The structure helps you stay oriented.
- Multi-agent by design. Different agents excel at different tasks. Goalward lets you pick the right tool for each task without switching apps.
- Verification is explicit. Task acceptance and goal verification are separate from process completion. You review evidence and decide what counts as done.
- Local execution only. Goalward is a desktop app that runs agents on your machine. There's no hosted backend, no cloud sync, no intermediary.
Planned improvements (not in priority order):
- Web demo: Browser-based preview with sample data on GitHub Pages
- Automatic retrospectives: Generate summaries and next-step suggestions from execution results
- Parallel execution: Run multiple agents on the same task using git worktrees
- Additional platforms: Windows and Linux builds
- Plugin system: Custom runtime adapters and integrations
See the product proposal for more context on future directions.
Workspace data is stored locally:
~/Library/Application Support/dev.goalward.desktop/
Goalward does not send data externally. Your agent credentials and code stay on your machine.
Contributions are welcome! Please read:
- CONTRIBUTING.md — Development setup and PR workflow
- CODE_OF_CONDUCT.md — Community guidelines
- SECURITY.md — Reporting vulnerabilities
MIT © 2026 Beace