The End of Linear Vibe Coding with Thurbox
Most developers treat their terminal like a single-lane highway, there is nothing wrong with that, I do too. You pick one tool, one context, one agent, and you drive it until you crash or finish. If you want to switch from debugging a backend service in Claude Code to refactoring a frontend component in Aider, you stop one, start the other, and lose your momentum. It’s a serial process in a parallel world.
We have multi-core CPUs and distributed systems, yet we still code linearly. That bottleneck isn’t just annoying; it’s expensive. It kills flow state. It forces you to context-switch manually, which is where mistakes happen.
Thurbox changes how you interact with your AI agents. It is not another agent. It is not a model wrapper. It is a persistent multiplexer for your coding workforce.
Imagine having Claude Code, Codex, Antigravity, and Aider running side-by-side in separate panes of a single terminal window. They are all working simultaneously. You are not waiting for one to finish before starting the next. You are conducting an orchestra, not playing a solo instrument.
Thurbox is for developers who are tired of waiting. It is for those who want to leverage the full power of multiple AI models simultaneously without losing their mind to context switching. It turns your terminal into a command center. It respects your workflow, preserves your progress, and scales with your ambition. It is not magic. It is engineering applied to the problem of attention. And in a world of noisy AI tools, silence and persistence are the ultimate luxuries.
Persistence as a Feature
What makes this distinct is its persistence. In standard setups, if your terminal crashes or you reboot your machine, your agent sessions die. You lose the context. You lose the half-finished refactor. Thurbox sits behind a tmux session that survives everything. You can quit the Thurbox interface, go home, sleep, come back, and your agents are still running. They are still thinking. They are still writing code.
This decouples your attention from the agent’s execution time. You launch a heavy refactoring task on a large codebase, close your laptop, and when you return, the work is done. It transforms AI from a chat partner into a background worker that respects your time.
The architecture is deliberately agent-neutral. Thurbox does not care if you are using Anthropic, OpenAI, or a local model. It launches the vendor CLI unmodified. This is crucial because it means you get new features the day they ship. If Claude releases a new tool-use capability, you don’t wait for Thurbox to update.
You just update Claude. Thurbox is just the container. It knows nothing about prompts or models. It only knows how to manage terminals. This simplicity is its strength. It avoids the bloat of trying to be everything to everyone. It is a lean, focused tool for managing complexity.
Capabilities & Features
- Parallel Execution: Run multiple agents (Claude, Codex, Aider, OpenCode etc.) side-by-side in one terminal.
- Crash-Proof Persistence: Sessions survive reboots and crashes via tmux/psmux backends.
- Git Worktree Isolation: Each agent gets its own worktree on a shared branch, preventing file conflicts.
- Remote & WSL Support: Manage SSH and WSL sessions from a single local TUI.
- Lua-Based UI: Fully customizable interface; every pane and keybinding is a editable Lua file.
- Global Screen Search: Find sessions by name, repo, or even the error text currently on screen.
- Headless CLI: Scriptable session management for automation and CI/CD pipelines.
- Inter-Agent Mailbox: Agents can pass data to each other without scraping terminal output.
- Forking Trees: Branch conversations into lead/worker hierarchies for complex tasks.
- Themes
- Automations and tasks: scheduled agent runs and a todo list whose items can be handed to an agent.
- Extensions
Workflow and Installation: Best for CI/CD
Using it is straightforward. On Linux or macOS, you run a single curl command to install both the TUI and the headless CLI. Windows users have a PowerShell equivalent. You need tmux and git. Once installed, running thurbox seeds your configuration directory and draws the interface. The left pane lists your sessions. The right pane is the live agent terminal. Pressing Ctrl+N opens the repo picker.
You select a repository, choose whether to use a worktree, name the session, and pick an agent. That’s it. The agent starts. You talk to it. Press Ctrl+N again to spawn a second agent on a different branch. Switch between them with Ctrl+J and Ctrl+K. The one you aren’t looking at keeps working.
If you need to find a specific session among dozens, Ctrl+/ triggers a global search. It searches not just by name but by the text on the screen. You can find a session by the error message it is currently displaying. This is invaluable when you have multiple agents debugging different issues.
You can also fork conversations, creating lead/worker trees where a main agent delegates tasks to sub-agents. For automation, there is a headless CLI that can create and drive sessions from scripts. This allows you to wire up a fleet of agents for CI/CD pipelines or scheduled maintenance tasks.
License
MIT Licensed
