nodal-agentos

agent
Security Audit
Warn
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Agent OS for Claude Code: a desktop app to plan tasks, hand them to Claude Code agents and follow their runs across all your projects. macOS, early alpha.

README.md

Nodal

CI
Release
Platform: macOS
License: MIT

Agent OS for Claude. Plan tasks on a board, hand them to Claude Code agents and follow every run across all your repos.

Español

Nodal board

Early alpha, macOS only. Independent project, not affiliated with Anthropic.

Install

  1. Download the latest .dmg from Releases (Apple Silicon and Intel) and drag Nodal to Applications.

  2. Nodal is not notarized by Apple, so macOS blocks the first launch. Run once:

    xattr -dr com.apple.quarantine /Applications/Nodal.app
    

After that, Nodal updates itself: it checks for a new version on launch and once a day (and from Nodal → Check for Updates… or Settings → Updates), shows the release notes and installs it only if you choose Update. Updates don't need the xattr step again. Nodal 0.1.0 predates in-app updates, so from 0.1.0 download the next version by hand one last time.

You need Claude Code installed and logged in, and git. gh is optional, for tasks that open pull requests. Open claude once in each repo and accept the trust dialog before running tasks there.

What it does

  • Board of tasks grouped in projects, each project with one or more local repos. Plans in markdown, acceptance criteria, priority and labels.
  • Chats per project: talk to Claude about the whole project or one repo, approve its edits and commands from the app, and turn the conversation into tasks with one click. Sessions show up in claude --resume.
  • Delegate each task to one of your agents (~/.claude/agents, the repo's or a plugin), a workflow, or plain Claude.
  • Isolated by default: every task gets its own git worktree and branch, so several run on the same repo at once.
  • Finish with uncommitted changes, a commit, or a pull request. Local work merges back into its base branch from the task panel, without touching your checkout.
  • Automatic review against the acceptance criteria: pass goes to In Review, fail to Blocked with the findings.
  • One queue with a global concurrency limit, and a live view of each run: phases, subagents, transcript, tokens and diff.
  • Linear, optional: import issues, route them to repos and keep statuses in sync. More task managers are planned.

How it works

Nodal drives the claude CLI you already have (claude --bg, --agent, /<workflow>) and reads progress from claude agents --json and the session files under ~/.claude/projects. Those files are an undocumented internal format, so a Claude Code update can break the parsing.

Everything stays on your Mac: an SQLite database, worktrees under ~/.nodal, API keys in the Keychain. No network server, no telemetry.

Agents (MCP)

While Nodal is open, agents can list projects and tasks, create and update tasks and read run results through MCP. The app listens on a Unix socket in its data folder (readable only by you, no network port) and nodal-mcp, which ships inside the app, bridges it to stdio. Register it:

claude mcp add nodal -- /Applications/Nodal.app/Contents/MacOS/nodal-mcp

From source, build it with cargo build --release --bin nodal-mcp in src-tauri (after pnpm build) and register src-tauri/target/release/nodal-mcp instead.

Tools: list_projects, list_tasks, get_task, create_task, update_task, list_executors, launch_run, review_task, cancel_run, get_run, list_runs and get_queue. Nodal's own chats launch it with --chat, which adds propose_task: the chat shows a task card and nothing is created until you accept it. A debug build of nodal-mcp talks to a debug build of the app.

The nodal-tasks skill teaches agents when to create a task instead of doing the work, how to write its plan and acceptance criteria, how to pick the repo and the executor, and how to read a run's result. Install it with skills:

npx skills add Mazp17/nodal-agentos --skill nodal-tasks

Build from source

Needs Node.js 20+, pnpm 10 and stable Rust.

git clone https://github.com/Mazp17/nodal-agentos.git nodal
cd nodal
pnpm install
pnpm tauri dev

Built with Tauri 2 (Rust + SQLite) and React 19 + TypeScript. See CONTRIBUTING.md to work on it and RELEASING.md for how versions ship.

License

MIT © Nodal contributors

Reviews (0)

No results found