Claws Out Summit | Kye Gomez - ClawSwarm
Session summary
The presenter introduces ClawSwarm as a natively multi‑agent system built on top of the Swarms Python framework, positioned as a simpler and slightly superior version of OpenClaw. Tasks are decomposed across specialized worker agents—for example, agents configured to search, trade, or launch tokens—while a director agent routes tasks to the most relevant workers. The presenter emphasizes that developers can easily define agents, name them, set autonomy levels, and attach tools and servers such as MCP and x42.
ClawSwarm is described as a lightweight, early‑stage project intended to demonstrate how something like OpenClaw can be reproduced with relatively little code. The presenter explains that the system ingests inputs from messaging applications including Telegram and Slack using gRPC rather than WebSockets, which they claim avoids security issues they associate with OpenClaw’s WebSocket usage. Installation is shown via pip, with a gRPC server and messaging gateway started from the repository. A Rust package and an in‑progress TypeScript version are mentioned for performance‑sensitive or alternative language use cases.
Key takeaways
- The presenter describes ClawSwarm as a natively multi‑agent architecture in which a director agent delegates tasks to specialized worker agents that can search, trade, or launch tokens.
- According to the presenter, ClawSwarm is built mainly in Python on the Swarms framework, allowing developers to configure named, autonomous agents and attach tools and servers such as MCP and x42.
- The presenter states that ClawSwarm ingests data from messaging platforms like Telegram and Slack via gRPC instead of WebSockets, which they argue reduces security issues they associate with OpenClaw’s WebSocket approach.