RxCode launches a workspace for managing teams of AI agents

A workspace for agent teams

RxCode is a native workspace designed for teams that use AI agents to manage software development workflows. Unlike standard chat interfaces, it organizes work into sprint boards where humans and agents share the same priorities, milestones, and versions. The platform is available on Mac, servers, and mobile devices, allowing for a synchronized experience across different hardware.

The core concept is a shared environment where tasks are represented as cards. When a user drops a card into the 'In Progress' column, an agent automatically picks it up to begin the work. This automation allows for a continuous development cycle where the software handles the movement of tasks between various stages of the pipeline.

Every run performed by an agent streams into its own dedicated thread. This transparency allows human developers to step in at any point to review progress or provide feedback. Once an agent finishes a task, the work lands in a review section with a summary of the changes, ready for final approval before it moves to the 'Done' column.

Automatic routing and cost control

One of the primary features of RxCode is its ability to perform automatic routing for tasks. The system reads the scope of each card to determine which specific agent and model are best suited for the job. This ensures that quick fixes are handled by fast, inexpensive models while deep refactors are assigned to more powerful 'heavy hitters.'

By matching the task complexity to the appropriate model, the tool helps teams save on token costs without sacrificing the quality of the output. This intelligent distribution means that developers do not have to manually select which AI to use for every individual piece of code or documentation. The system handles the logistical overhead of agent selection automatically.

The platform also supports a wide variety of existing tools and models through the Agent Client Protocol (ACP) registry. Users can install various runtimes, such as Cursor, Gemini CLI, and GitHub Copilot, into the same chat and board system. This flexibility allows teams to mix and match different AI capabilities within a single unified workspace.

Mobile sync and remote oversight

RxCode is built to function as a mobile-paired workspace, meaning users can manage their agent teams from an iPhone, iPad, or Android device. The mobile app allows users to follow live sessions, preview development servers, and approve specific tool calls as they happen. This enables a 'work from anywhere' model where the heavy lifting happens on a server while the human provides oversight on the go.

Security is a central component of this remote connectivity, as the system uses end-to-end encryption for its relay. The relay only forwards ciphertext, ensuring that sessions stay private between the paired devices. Users receive push notifications for finished runs, briefings, and tool call approvals, allowing them to tap a button to grant permission or deny a request.

The mobile experience is designed to be a seamless extension of the desktop or server environment. A user can start a complex task on their Mac, hand it off to the agents to continue while they are away, and then receive a morning briefing on their phone. This transition allows for a continuous workflow that isn't tethered to a single workstation.

Local fine-tuning and privacy

For users concerned about data privacy, RxCode includes a feature for local fine-tuning. This allows the system to train a lightweight adapter on the work the user has already completed, such as previous threads, diffs, and task reviews. Because this happens on the user's machine, nothing is uploaded to the cloud during the training process.

Once a model is fine-tuned locally, it can be used to process routine tasks in the user's specific style. This is particularly useful for offline work, as the tuned model can pick up tasks with zero API costs and no requirement for a network connection. It effectively creates a personalized AI assistant that understands the user's unique coding patterns and preferences.

The platform emphasizes a private-by-default approach, where secrets are encrypted on-device using passkeys. This ensures that the cloud workspace never sees plaintext data. By combining local training with a shared cloud workspace for team collaboration, the tool attempts to balance the need for collective productivity with individual data security.

Autonomous workflows and reporting

RxCode is designed to keep agents 'on call' 24 hours a day to handle repetitive maintenance tasks. While a human developer is asleep, the agents can watch CI (continuous integration) pipelines, triage new issues, and review dependency updates. This allows the team to maintain a steady pace of development without requiring constant human intervention for every minor bug or update.

The system can turn simple prompts into recurring tasks, such as performing issue triage every two hours or drafting release notes every Friday. For example, an agent can be tasked with drafting a morning summary before a developer starts their day. These summaries are then sent to the user's phone, providing a clear overview of what was shipped, fixed, or decided overnight.

The tool also facilitates communication between agents across different projects. A user can ask the system questions about their board, briefings, or codebase, and the agents can interact with each other to resolve blockers. This creates a multi-agent ecosystem where the AI handles the coordination of small tasks, leaving the human to focus on high-level planning and final approvals.