Murmell

Product Introduction

1. Definition: Murmell is a cloud-based, collaborative development environment (CDE) and multi-agent orchestration platform. It provides a unified browser canvas where multiple AI coding agents (like Claude Code, OpenAI Codex, and Kimi) can operate simultaneously on a single, shared project directory hosted on a dedicated cloud sandbox.
2. Core Value Proposition: It exists to solve the fundamental coordination and infrastructure problems of running multiple AI coding agents in parallel. Its primary value is enabling real-time, conflict-free collaboration between AI agents and human developers by enforcing file-level locking at the system level, preventing merge conflicts before they happen, and providing a persistent, stateful workspace that survives disconnections.

Main Features

1. Unified Agent Memory & Persistent Workspaces: The platform maintains a continuous state for the entire development environment. When a user closes their browser or laptop, the canvas and all agent sessions are frozen. Upon returning, the workspace—including open terminals, file states, and agent conversation contexts—is restored exactly as it was. This is powered by cloud-based session persistence and container snapshotting technology.
2. Real-Time File Path Reservation System: This is the core technical innovation. Before an AI agent writes to a file, it must claim a reservation for that specific file path. A central "board" enforces a one-holder-per-path rule. This server-side concurrency control prevents overlapping writes and destructive edits by other agents, a common failure mode in multi-agent coding. Conflicting writes are automatically snapshotted for recovery.
3. Live Multiplayer Collaboration Canvas: The interface is a shared, real-time visual workspace. Multiple human users can join the same canvas via a link to see live cursors, active agent terminals, and a synchronized live preview of the application. Each user can independently control their own set of connected AI agents within the same shared project environment, facilitated by WebSocket-based real-time communication.
4. Multi-Agent Orchestration & Management: Users can spawn and manage several different AI coding agents (Claude, Codex, Kimi, OpenCode) from a single dashboard. Each agent runs in its own isolated terminal window on the shared cloud machine. Tasks can be described once and delegated across agents, who then coordinate via the file reservation system to work on different parts of the codebase concurrently.
5. Integrated Private GitHub Repository: Every Murmell project is automatically linked to a private GitHub repository owned by the user. The cloud workspace acts as the working directory, and code is automatically committed and pushed to this repository, ensuring the work is preserved independently of the ephemeral cloud sandbox machine.

Problems Solved

1. Pain Point: Agent-Over-Agent Conflict. When running multiple AI coding agents locally or in separate sessions, they frequently overwrite each other's changes, leading to broken code, lost work, and manual merge conflict resolution, negating the efficiency gains of using AI.
2. Target Audience: Development Teams and Solo Developers leveraging AI coding assistants for complex projects; Tech Leads orchestrating work across multiple AI models; Startups and Agencies looking to parallelize development work and enable real-time collaborative coding reviews with AI.
3. Use Cases: Rapid Feature Development: Spawning specialized agents (one for backend, one for frontend, one for CSS) to build a feature module simultaneously. Live Collaborative Debugging: A team lead and an engineer joining the same canvas to watch an AI agent debug an issue in real-time, discussing and guiding the process. Persistent Long-Running Tasks: Setting up an agent to refactor a large codebase over several hours or days, with the ability to safely pause, disconnect, and resume monitoring from any device.

Unique Advantages

1. Differentiation: Unlike standalone AI coding tools (Cursor, GitHub Copilot) or cloud IDEs (GitHub Codespaces, Gitpod), Murmell is not just an editor with AI. It is an orchestration layer and concurrency manager specifically designed for the multi-agent paradigm. Competitors focus on a single AI user experience, while Murmell focuses on managing the interactions between multiple AI actors.
2. Key Innovation: The server-enforced file reservation system. Instead of relying on AI prompts or post-hoc git merges to manage conflicts, Murmell bakes coordination into the infrastructure. This deterministic, system-level lock ensures data integrity, making parallel AI coding predictable and reliable—a significant technical advancement over ad-hoc coordination attempts.

Frequently Asked Questions (FAQ)

1. How does Murmell prevent AI agents from overwriting each other's code? Murmell uses a real-time file path reservation system. Before any agent can edit a file (e.g., src/App.tsx), it must claim a lock on that specific path. This lock is enforced by Murmell's central server ("the board"). If another agent attempts to edit the same file, its request is queued until the lock is released, completely preventing overwrites and merge conflicts at the source.
2. What happens to my Murmell workspace if I close my browser? Your workspace is preserved in its exact state. Murmell's cloud canvas is persistent; it freezes the entire environment—including running agent processes, terminal output, and open file states. When you return, the canvas resumes from where it stopped. This "unified agent memory" means AI agents retain their full context and can continue tasks seamlessly.
3. Do I pay Murmell for AI model usage (tokens)? No. Murmell does not resell AI model tokens. You connect your own existing accounts and API keys for services like Anthropic Claude, OpenAI Codex, or Moonshot Kimi. Murmell's subscription fee covers the cost of the managed cloud infrastructure, the collaboration platform, and the concurrency control technology.
4. Is my code secure and isolated on Murmell's cloud? Yes. Each project/canvas runs on its own isolated cloud machine (sandbox). These execution environments are ephemeral and contain no persistent Murmell database or decryption keys. All final code is automatically pushed to a private GitHub repository under your control, which serves as the source of truth.
5. Can my entire team work in the same Murmell canvas simultaneously? Yes, this is a core feature. On the Pro and Builder plans, multiple users can join the same canvas via a shareable link. All participants see the same live view: active agent terminals, user cursors, and previews. Each user can spawn and command their own agents within the shared project, enabling true real-time, multi-user, multi-agent collaboration.