DOCSEARCHbyAlgolia
Submit Docs

CubePlex Open Source Agent Workspace

website·Indexed 30 days ago

https://docs.cubeplex.ai/

An open-source agent environment and workspace designed for teams to maintain repeatable work. CubePlex provides a framework where agent activities are visible, hand-off capable, and reusable within a shared Workspace. It includes CubeSandbox, which utilizes a KVM MicroVM execution stack to handle high-concurrency creation, state preservation, and rapid recovery for isolated code execution.

Tokens
124.9K
Snippets
270
Records
883
Agent score
48%

What's inside CubePlex

  1. Overview of the CubePlex isolated sandbox

    CubePlex executes agent-generated code in an isolated sandbox powered by OpenSandbox. The agent interacts with the sandbox via shell-based tools to issue commands, write files, and run scripts. The available languages and tools (e.g., Python, Node.js) are determined by the default_image configured in the organization's sandbox policy. Commands taking longer than 60 seconds are treated as durable background tasks with a default execution deadline of one hour.

  2. Overview of YC Quartermaster (QM)

    Quartermaster (QM) is an open-source, self-hosted multiplayer agent harness developed by YC. It organizes people, Slack channels, groups, and projects into 'scopes,' where each scope serves as a distinct agent work environment and a permission boundary.

  3. Overview of CubePlex Team Agent Workspace

    CubePlex is an open-source AI Agent workspace designed for teams to ensure that Agent-driven work is visible, transferable, and reusable. Unlike individual chat interfaces, CubePlex organizes conversations, members, Skills, Memory, MCP tools, and isolated execution environments into a shared Workspace. This prevents project knowledge and tool configurations from being siloed within a single user's chat history.

  4. Overview of QM (Quartermaster) for organizational agent management

    QM (Quartermaster), open-sourced by YC in July 2026, is a framework designed to manage fleets of AI agents across an organization. It addresses the difficulty of managing numerous personal agents by providing centralized control over models, permissions, and the runtime while maintaining individual agent flexibility. QM implements scope-based collaboration, where interactions are isolated to specific contexts—such as Slack direct messages, channels, group DMs, and projects—preventing the need to share a single user's entire environment with the whole company.

  5. Overview of CubeLoop Agent Framework

    CubeLoop is a high-performance, async-native Python agent framework designed for production use. It focuses on traceability and includes production-grade persistence to ensure agent activities are recorded and reproducible.

  6. Overview of CubePlex Team Agent Workspace

    CubePlex is an open-source Agent Workspace designed for teams. It utilizes a model where long-term Workspace Agents are defined first, and the specific session, identity, and execution boundaries for a run are determined by the entry point used (such as Web, Slack, Discord, Microsoft Teams, Lark/Feishu, DingTalk, or WeChat Work). This architecture allows a single team Agent to maintain continuity across different entry points while ensuring that private chats, authorizations, and execution environments remain isolated and do not interfere with one another.

  7. Overview of CubePlex Sandbox Execution

    CubePlex uses OpenSandbox to execute agent-generated code in an isolated environment. This allows agents to run code, install packages, and interact with files without affecting the host infrastructure. Agents execute commands via a shell tool, meaning they are not limited to a specific programming language.

    Key execution behaviors:

    • Streaming Output: Command output is streamed in real-time to the response.
    • Background Tasks: Commands running longer than 60 seconds (excluding sleep) become persistent background tasks to allow the current response to complete.
    • Timeouts: Tasks use a configured command deadline (default is one hour). Long-running commands are automatically terminated to prevent resource exhaustion.
    • Tooling: Available languages and tools (e.g., Python, Node.js) are determined by the sandbox image configured as the default_image in the sandbox policy.

  8. Overview of QM (Quartermaster) agent harness

    QM (Quartermaster) is an open-source, self-hosted multiplayer agent harness developed by YC. It organizes people, Slack channels, groups, and projects into 'scopes,' where each scope serves as both an agent work environment and a permission boundary. This model is designed for operating multiple independent agent computers within a company's existing permission and runtime system.

  9. Overview of CubePlex Discord Integration

    CubePlex integrates with Discord as a gateway bot, establishing a persistent outbound socket connection to receive messages. This architecture means the CubePlex host does not need to be exposed to the internet. The integration supports native Discord slash commands (/new, /reset, /link) which appear in the Discord command selector.

  10. Overview of CubePlex Event Triggers

    Event triggers allow you to start agent runs when external events occur via HTTP POST webhooks. CubePlex provides a unique webhook URL and an HMAC signing secret for each trigger. Because CubePlex uses a proprietary signing scheme, it does not natively accept third-party signatures (e.g., from GitHub or Stripe). To integrate these services, you must use a backend or relay to re-sign the payload according to CubePlex's requirements.

  11. Overview of YC Quartermaster (QM)

    Quartermaster (QM) is an open-source tool by Y Combinator designed for organizational agent management. It provides a framework to manage models, permissions, and runtimes while maintaining the flexibility of personal agents. QM implements a scoped collaboration model where Slack direct messages, channels, group DMs, and projects resolve to different scopes, preventing the need to share a single person's environment with the entire company.

  12. Overview of CubePlex Event Triggers

    Event Triggers allow external events to start an agent run in CubePlex. When a trigger is created, CubePlex provides a unique webhook URL and a signing secret. Any service capable of sending an HMAC-signed HTTP POST request can trigger the agent. Note that CubePlex does not natively support third-party signature formats (like GitHub's X-Hub-Signature-256); therefore, you must use your own backend or a relay service to sign requests according to the CubePlex scheme.