YC Just Open-Sourced Your AI Team

Y Combinator's new open-source agent, QM, represents a significant architectural shift from the typical agent "harnesses" flooding the ecosystem. This isn't another ephemeral terminal utility designed for local execution, like Claude Code. QM moves beyond individual developer tools, establishing a centralized, organizational service that integrates deeply into company workflows.

Instead of a CLI, QM ships as a full-stack web application, built for self-hosting directly on your company's infrastructure. This design choice means organizations deploy a single, robust instance, with a headless backend managing agent loops and session history in Postgres, entirely off-device. It’s a foundational service, not a desktop app, allowing every team member to leverage a consistent, shared AI platform.

Crucially, QM provides each user with an isolated, stateful workspace. These aren't temporary sessions; they are containerized environments, each with dedicated storage that preserves installed packages, cloned repositories, and all generated files across multiple sessions. Imagine setting up a complex Python environment on Monday and finding it perfectly intact and ready for use on Friday – embodying true operational persistence.

Accessing QM is seamless, whether through its intuitive web UI or directly within Slack, integrating the agent where your team already works. All agent loops and tool calls execute securely in the cloud, ensuring your local machine remains unburdened and freeing resources. This ubiquitous access, combined with cloud-based, persistent execution, delivers a powerful, always-on AI team member for your organization, ready for the future of collaborative automation.

QM’s multiplayer design fundamentally redefines team interaction with AI. Instead of isolated local agents, it anchors on shared projects, allowing collaborative teams to pool resources. Users can share memory, files, applications, and credentials, establishing a centralized knowledge base that compounds organizational intelligence. This means teams can set up tools and data once, and everyone benefits instantly.

This architecture thoughtfully balances individual autonomy with collective needs. Each user maintains a distinct personal space for private configurations, including sensitive credentials. Simultaneously, they contribute to and access a shared project context, fostering a collaborative environment without compromising security or personal control. This duality, crucial for enterprise adoption, ensures that individual contributions seamlessly integrate into a shared pool of resources.

While its current web UI falls short in clearly indicating the active context—a common early-stage challenge—the foundational architecture robustly supports team-based workflows. This design directly solves the pervasive isolation problem found in single-user agent setups. QM builds a cohesive environment where an organization's collective work and knowledge are amplified, moving beyond individual silos to a truly shared operational layer.

QM doesn't force a single approach. It embraces flexibility by allowing organizations to Bring Your Own Harness. For instance, familiar harnesses like Claude Code can run seamlessly within QM’s managed cloud environment, offering the best of both worlds: established agent logic combined with QM's centralized, stateful infrastructure. Teams aren't locked into new paradigms but can integrate existing workflows.

Simplicity defines QM's agent interaction model. Its minimalist tool surface provides agents with just seven core tools: execute, read, write, publish, memory, history, and background. This design choice forces agents to leverage shell commands within their isolated sandbox, simplifying agent design and interaction by mirroring human-like machine operation rather than complex API wiring.

Crucially, QM addresses enterprise concerns with robust safety guardrails. Organizations can configure three org-wide security postures:

• Strict: Requires human approval for nearly every tool call.
• Auto: The default, employs a classifier for external data and tool results.
• Dangerous: No screening or pauses, for high-trust scenarios.

A vital command policy also blocks destructive actions, like recursive deletes or destructive SQL, even in "Dangerous" mode, ensuring fundamental safety for production environments. For more technical details on its architecture and features, refer to the project's GitHub repo: yc-software/qm: Multiplayer agent harness for work. https://qm.ycombinator.com.

QM's core value proposition shines through: it is open-source, self-hostable, and massively customizable. This full-stack web application fundamentally re-architects agent workflows from fragmented, single-player instances into a cohesive, shared organizational service. It centralizes agent memory, files, applications, and credentials, finally enabling true collaborative AI within teams.

Yet, this vision comes with early-stage caveats. The project is extremely young, merely two weeks old upon its initial release. Its UI/UX is demonstrably immature; for instance, the web UI often lacks persistent indicators for the active context. Crucially, configuring new harnesses, like integrating Claude Code, currently bypasses the admin panel entirely, demanding direct API calls to the /harnesses endpoint.

QM is not a polished, plug-and-play solution. Its current state positions it as a significant opportunity for technical teams ready to manage deployment, contribute to its evolving codebase, and embrace its early-stage complexities. For those willing to invest, QM offers a powerful, foundational glimpse into a more integrated, multiplayer AI future, where agents operate as shared, persistent resources rather than isolated tools.

What is YC's QM?

YC's QM is an open-source, multiplayer agent harness. It's a self-hosted web application that provides teams with collaborative, persistent, cloud-based environments for running AI agents.

How is QM different from local agent tools like Claude Code?

While QM can run harnesses like Claude Code, it's a full-stack platform, not a local CLI tool. It offers persistent cloud workspaces, shared context for teams, and centralized management and safety controls.

Is QM ready for production use?

QM is very new and its UI/UX is still immature. While powerful, it's best suited for technical teams comfortable with self-hosting, API interactions, and contributing to an early-stage open-source project.

What are the main benefits of QM for a team?

The key benefits are shared state and context. Teams can collaborate on projects with shared files, memory, and tools, and work is done in persistent cloud environments accessible from anywhere.