Spool: Claude Code Agents for Telegram and Slack

What changed

Spool is a new open-source project designed to facilitate the creation, management, and interaction with persistent Claude Code agent loops. The system allows these loops to coordinate their work, communicate directly with each other, and remain accessible through chat interfaces like Telegram and Slack. A key feature is its operation on the user's own Claude subscription, bypassing the need for direct API costs and instead utilizing the user's existing plan terms. The project is actively developed in the open, with its own development process involving Claude Code loops building the system itself, managed through pull requests and code reviews.

Key Features and Functionality:

• Persistent Loops: Agents can maintain their state and continue work across sessions, with scheduled wakes and resumable tasks.
• Direct Coordination: Loops can communicate directly within a fleet channel, reducing the need for manual message relaying by the operator.
• Chat Integration: Interaction with loops is possible through a web control room, Telegram, or Slack, allowing users to stay involved without direct terminal access.
• Self-Hosted Operation: Spool runs on the user's infrastructure, utilizing their Claude subscription for execution. This includes options for contained Docker environments, uncontained bare processes, or cloud sessions managed by Anthropic.
• Plan Guardrails: The system monitors usage against Claude plan limits (e.g., 5-hour and 7-day windows) and can automatically pause loops to prevent exceeding caps, with options to resume or adjust thresholds.

Setup and Runtime:

To run Spool, users require Linux, macOS, or Windows with WSL2, along with Docker, Go (version 1.25+), and Node.js (version 20.19+ or 22.12+). Claude Code must be installed and accessible via a Claude subscription. The quick start guide involves cloning the repository, building the project, and starting the control room. Users then sign in with their Claude account, which Spool uses to harness the model's capabilities. New loops can be created with a name and mission, and chat surfaces like Telegram or Slack can be attached by configuring bots and providing tokens.

Why it matters for builders

Spool empowers builders by providing a framework to deploy sophisticated, multi-agent AI systems without the typical overhead of managing separate API endpoints and associated costs. The ability to run agents on one's own Claude subscription democratizes the use of advanced AI coordination. Developers can experiment with complex workflows, persistent agent memory, and seamless integration into existing communication channels, fostering more autonomous and collaborative AI applications.

Practical impact

Builders can leverage Spool to create autonomous agents for tasks such as code generation, content creation, or complex problem-solving, where multiple specialized agents can collaborate. The direct integration with Telegram and Slack means these AI coworkers can be part of daily team communications. The project's open-source nature and active development suggest a community-driven evolution, offering opportunities for contributions and customization. Developers can start by cloning the repository and following the quick-start guide to set up their first loops and integrate them with their preferred chat platforms.

Caveats and source limits

Spool is currently pre-1.0, indicating that its features and stability are subject to change. The source material does not provide independent benchmark results for performance or cost-efficiency compared to other agent frameworks. While it mentions operating on the user's Claude plan, specific details on how plan limits are enforced or potential throttling by Anthropic are not elaborated upon. The security section outlines containment measures, but a full security audit or independent verification is not mentioned. The project's reliance on the Claude Code model means its capabilities are tied to that specific AI, and no information is provided on support for other LLMs.