chat with agentaskProjects2 folders 4 filesAI and agents2 folders 2 filesInfrastructure3 foldersDevelopment5 filesAbout5 files
Helm: a console for parallel agents
  • Svelte 5
  • TypeScript
  • Fastify
  • node-pty
  • xterm.js
  • tmux
  • SQLite
  • WebAuthn

Helm: a console for parallel agents

A console for running several Claude Code sessions at once: a real terminal in the browser, session statuses, limit analytics, memory and tasks. It works from a phone too.

2026 · author

Screenshots

  • Helm main screen: parallel agent sessions with statuses on the left, the selected session terminal on the right, limits and machine load at the top
    Sessions and terminal
  • A file from the agent output opened in the side panel next to the terminal, with syntax highlighting and an edit button
    File viewer opened from the terminal
  • Analytics screen: 5-hour and weekly limits with a forecast, time by project, agent wait times and an hourly chart
    Analytics and limits
  • Memory screen: a graph of knowledge base notes linking projects to decisions, lessons, plans and journal entries
    Memory: notes graph
  • Tasks panel: new threads from the journal, personal tasks and open tasks for each project
    Tasks by project
  • Mobile readable view of a session: agent messages, collapsed actions, a table and the agent's question with a Reply button
    Session chat on the phone

I usually run several Claude Code sessions in parallel, and watching them across a pile of terminal windows is awkward: it is hard to tell who has finished, who is waiting for an answer and who is burning through the limit. I tried the existing consoles, none of them fit, so I built my own.

Helm lists all active sessions in a sidebar with statuses from hooks: working, waiting for permission, asking a question, done. On the right is a real terminal: xterm.js in the browser connected to tmux on the Mac, with plain claude inside. Files the agent mentions in its output open in a side panel.

Separate screens cover limit and time analytics (how long I worked, how long the agents did, where I held them up), the memory graph and feed, and tasks per project. On a phone Helm opens as a PWA with Face ID sign-in: a chat instead of the terminal, button replies and push notifications when an agent is waiting for me and the Mac is idle.

269 commits and over 1,500 tests in twelve days. All screenshots use demo data.