Multiple accounts. One workspace.
Add and organize accounts for your agents and check each account’s sign-in status in one workspace.
The Programming Terminator
Bring your existing coding agents and accounts into one development workspace. Switch accounts within the same agent while keeping the conversation, or hand work to an agent from another provider. Choose an agent to coordinate tasks across your projects.
macOS on Apple silicon and Intel · Windows on x64 · Linux server on x86_64 and aarch64 · Beta
Accounts, context and collaboration: find the tools that remove friction from your next task.
Illustrations with sample data
Add and organize accounts for your agents and check each account’s sign-in status in one workspace.
Switch accounts of the same agent and continue the conversation, manually or with automatic switching at the usage limit.
Automatically pick an available account for new sessions. With automatic switching on, a task with no eligible account waits and resumes after a fresh reading confirms recovery.
View account usage and reset countdowns together, with readings refreshed automatically.
Accounts of the same agent can share skills, plugins and MCP servers. Claude Code’s user-level MCP servers stay separate. On Windows, enable Developer Mode on the daemon host.
Check context usage when the agent reports it and start the next stage in a fresh session. Revisit the earlier conversation while its record is available.
Let an agent from another provider take over from a handoff record of the task, progress and remaining work.
Open a stopped session in its agent’s app on the Mac or Windows PC running the daemon.
Give one request to a coordinator that assigns work to the agents for each project, follows progress and brings their reports together.
Work in separate native CLI sessions that load each project’s instructions, skills, hooks and MCP configuration.
See which sessions are working or waiting, then open the project session to answer a question or adjust instructions directly.
See processes started by agents, their memory, listening ports and local web links in real time. When a session ends, Octoboard tries for a limited time to stop the processes it tracked and you haven’t chosen to keep.
Browse project files and Git diffs beside the session to review what the agents delivered.
Try the workspace with sample projects. Switch between the project overview, focus mode, and code review.
Actual application UI · Sample data
Take a meal promotion across a restaurant’s ordering, payment and store-management projects. The illustration below shows a coordinator assigning tasks and receiving reports. Select a project to see its assigned task and example report.
Add a meal deal promotion across our restaurant apps.
CoordinatorThe coordinator assigns work, coordinates shared promotion rules and API changes, and collects the results.
Project instructions · Skills · Hooks
Build controls for meal deal prices, promotion dates, and eligible locations.
Administrators can set meal deal prices, promotion dates, and eligible locations. The promotion rules are shared with the other projects.
Your coordinator uses the goal and project context to identify the projects, break down the work, and track dependencies.
Each agent works in its own project directory, using that project’s instructions, Skills, hooks, tools, and permissions.
Project agents send progress and results to the coordinator. Open any project’s conversation to add context, adjust direction, or work directly with its agent.
Use the macOS or Windows app to work on a Linux server or another computer through SSH, an encrypted connection straight to that machine, with no relay or cloud service.
I built Octoboard because I believe people no longer need to write code by hand. Developers can put their attention back where it matters: the product they want to deliver.
As AI produces more code, reviewing and rewriting every line by hand can no longer keep pace. I see the shift toward implementation happening inside a black box as irreversible. In my view, AI already writes more complete logic than most programmers. The conventions for “good code” that we designed around human readers are losing their central role as AI takes on both writing and maintenance.
I see the biggest obstacle in AI development as misunderstanding: unclear requirements, missing context, or assumptions left unstated. The work that deserves our attention is defining the problem, explaining constraints, and giving agents the context they need to work toward the same goal.
For me, software quality has one deciding standard: whether the product meets real needs under thorough testing. Reading code cannot establish that on its own. Tests need to cover real workflows, edge cases, and failures, and their results need to guide the next iteration.
That is the purpose of this workspace: coordinating agents, carrying context forward, and following the work through to delivery. As developers spend less time in code editors, they can devote more of it to requirements, decisions, testing, and the product itself.
If you also believe people no longer need to write code themselves, this workspace is built for you.
Octoboard stores its application data locally and does not upload your code or conversations to its own servers. Your chosen agents and tools connect to their own providers according to your configuration.
Explore how agents are coordinated, and adapt the code to the way you work.
The app is free and open source under the MIT License.
The macOS and Windows apps and the Linux server are in beta. Reach the Linux server from the desktop app or a browser over SSH. iOS and Android apps are planned.
Beta
Beta
Server · Beta
Planned
Planned
Requirements for each system, and a link to its installation guide, are on the download page. Download
Add your existing project folder without moving your code. The default account uses your existing CLI configuration, and the agent still loads your project's instructions, skills, hooks, and MCP servers. Additional accounts use separate configurations, with optional sharing of supported settings.
Octoboard keeps its application data locally and does not upload your code or conversations to its own servers. Claude Code, Codex and Grok Build use services you subscribe to or configure, with your own accounts; these tools manage account credentials and conversation records. They send prompts, code and relevant context to the providers you configure, under those providers’ policies. When agents collaborate, task descriptions, progress and relevant context also pass to the agents taking part.
Developing several projects in one session can mix their contexts. Working across directories or using subagents does not guarantee that the target project’s full configuration is loaded: rules can be missed when the parent session has to pass them on, and loading behavior varies by tool. Octoboard starts an independent native session of the chosen agent in each project directory, so the agent loads local configuration through its own mechanisms while the coordinator organizes the work.
Yes. Choose Claude Code, Codex or Grok Build independently for the coordinator and each project, based on the task and the project’s tools. Each agent keeps its own instructions and toolchain, while Octoboard passes tasks, relevant context and reports between sessions. The coordinator can delegate work and bring the results together.
A requirement or code detail may be easier to explain directly to the project agent doing the work. You can enter its conversation to add context, adjust requirements, or review code; sessions started by the coordinator still report results, open items, and decisions needed back to it.
Install and configure the coding-agent command-line tools you want to use first. Octoboard is free and open source under the MIT License; agents and model services follow the subscriptions or billing terms of the providers you choose. If your agent supports a self-hosted model, you can keep using that configuration too.
Yes. Run agents on your Linux server or on another Mac or Windows PC running Octoboard. Connect to that machine over SSH from the desktop app or a browser.
Sessions share the project folder by default, so simultaneous edits to the same file can conflict. For parallel work, ask the agents to use separate Git worktrees and review the changes before merging.
Yes. Project files stay in their original folders and each agent keeps its own conversation records. While the original agent still has a conversation record, you can resume that conversation through its command-line tool. With an extra account, start the tool using that account’s configuration directory.


The Programming Terminator
Programming ends here. Power to rival the Terminator.
Beta · MIT License