Flamerock: An App for Running Dozens of Claude Code Agents From One Screen
Overview
On a normal day I have a dozen or more Claude Code terminals open, on a work account and a personal one, each with its own agents and sub-agents running under it. The hard part was never the agents. It was me: which terminal is waiting for my answer, where was that question Claude asked an hour ago, which window was the fork of which session. I lost the thread a lot. So I built Flamerock, a local dashboard that sees every session at once, and it has been my main tool every day for weeks now.
One screen for every session
Flamerock lists every live Claude Code session and the whole history, for both accounts, grouped by project: sessions in Terminal, in VS Code and in Claude Desktop, with their sub-agents under them. Each row shows whether Claude is running, idle or waiting for me, and the ones waiting for me stand out. Sessions I care about most go to a Pinned list above the live ones.
Finding things is one key away: / opens the search, and every word I type narrows the sessions and the cards together, by title, first prompt, project or card text. Then one click on Go to terminal brings the right Terminal tab to the front, or the right VS Code window, and a closed session opens again from history with Open, where it left off. A rail on the left keeps my project folders, each with its own avatar and Terminal profile, and starts a new Claude session in the terminal or in VS Code with one click.
Answers become cards
Long answers hide the parts that need me: a question, a risk, a to do. Flamerock reads Claude's answers and turns those parts into cards, automatically or when I press Make cards on a session, and I can add my own with + Card. A card can start a fork that works on it, become a Notion task or a GitHub issue, or be pinned so it stays on top. Sessions get a status tag too (to do, start, important, done), so following up is a glance, not a hunt.
Forks that stay in order
Forking a conversation in Claude Code is easy; keeping track of the forks is not. In Flamerock a fork is one click, from a session or from a card, and it shows up under its parent in the list. When the fork is done it hands back: Flamerock asks it for a short report, I read and edit it, and it goes to the parent session, so the main conversation keeps the result without the fork's whole history.
A closed terminal is not a lost session either. The session detail shows the whole conversation of any session, open or closed: my prompts, Claude's answers and the tools it ran, including what the terminal hides.
Every row, my call
Each live row has four switches: pin it to the top, mute its cards, hide its Terminal window while Claude keeps working, and Remote, which turns on Claude's own Remote Control for that session, so I can carry on from my phone.
The phone is what I am building next. A small Flamerock view for the phone that answers one question, by voice if I like: who needs me? It lists the sessions waiting for me, and one tap opens the right one in Claude's Remote Control. The Mac stays the brain; the phone is only a screen and a microphone.
It also looks the way I like: ten themes, from my own findmutlu colours to a parchment look, and an avatar for every project folder.
How it is built, and what is next
Flamerock runs on my Mac as a small local server that starts at login and serves a React page. It reads Claude Code's own session files from both accounts, listens to Claude Code hooks for the moments a session stops or waits, and asks Claude Haiku to pick the cards out of an answer. It never types into a terminal: it only brings windows forward or opens new ones, and the only command it sends is Claude's official Remote Control. Everything stays on the machine.
Flamerock is not open source yet. I built it for myself and it carries my own settings, so I am cleaning it up into a general version first. It will be MIT licensed, and this page will link the code when it is out. Flamerock is an unofficial dashboard for Claude Code sessions, not a product of Anthropic.