Open Source Friday with Clawdbot 🦀

Key Concepts

• Democratization of AI: Clubbot aims to make personal AI agents accessible to everyone, regardless of technical expertise or financial resources.
• Open-Source & User Control: The project prioritizes open-source principles, granting users control over their data and the ability to run agents locally.
• Agent-Based Automation: Clubbot facilitates a shift from prompt engineering to proactive, automated task completion through intelligent agents.
• Hardware Accessibility: The project demonstrates the viability of running AI agents on older, repurposed hardware.
• Community-Driven Development: Clubbot relies heavily on community contributions for growth, maintenance, and innovation.

Project Origins & Architecture

Peter, motivated by burnout and the limitations of existing AI assistants, began developing Clubbot in November as a WhatsApp-based prototype. The project’s core motivation is to provide a customizable, open-source alternative to large AI labs, offering users data control and local execution. Initially a “gluing together” of existing services (WhatsApp, Cloud Code), Clubbot’s architecture now supports multiple messaging platforms (WhatsApp, Telegram, Discord) and image processing. It leverages models like OpenAI’s Opus and MiniMax’s M2.1, with a focus on local model compatibility. A key feature is the “heartbeat” mechanism, enabling proactive task checking and updates. Security is addressed through sandboxing and an allow-list system, though “YOLO Mode” exists for minimal restrictions. An unexpected breakthrough occurred when the system processed a voice message using an OpenAI key despite lacking explicit voice support, demonstrating its resourcefulness.

Community Growth & Adoption

Clubbot has rapidly gained traction, exceeding 600+ GitHub stars. The community is actively contributing diverse applications, including automated subtitle generation, Tesla integration, and public transport monitoring. Notably, many contributors are new to coding, using Clubbot as an entry point to open-source development. This growth has driven demand for Mac Minis as users set up local instances. Peter is managing the influx of pull requests by prioritizing the underlying problem being solved rather than code perfection, fostering inclusivity.

Hardware & Implementation

Running Clubbot on older hardware, such as 2013 iMacs, is a viable and environmentally friendly option. Installation is simplified with a one-liner command for the terminal. Hosting on affordable cloud providers like Hetzner, Fly, or X is also possible. A new system utilizing Gateway in the cloud with a local node connected via Tailscale is under development to access Mac-specific features like iMessage and the Photos app.

Model Selection & Performance

Entropic is currently considered the best-performing open-source model for Clubbot, with anticipated improvements. MiniMax offers a budget-friendly alternative, providing comparable request limits to OpenAI’s $100/month plan at $10/month, though it’s not yet on par with Entropic. Gemini models are explicitly deemed unsuitable for tool calling or the project’s intended use case, though potentially useful for code writing.

Contribution & Future Development

The project actively seeks contributions in documentation, bug reporting, Discord support, and testing. The developer acknowledges a significant workload (around 300 commits daily) and the need for community support. Planned releases include stable, beta, and “deaf” channels. Future development priorities include simplifying installation across systems, developing native apps for iOS, Android, and macOS, and enhancing the onboarding wizard with clearer security explanations and sandbox options. The goal is a user experience that allows easy control over bot access levels.

Real-World Applications & Anecdotes

Clubbot has demonstrated practical applications, including reducing user anxiety through task automation (like phone calls) and successfully checking the developer into a British Airways flight, accessing passport information. The vision extends to agent-to-agent communication for automated tasks, such as waking a user up.

Technical Considerations & Challenges

The project utilizes various technologies including Rust, TypeScript, Go, Swift, Kotlin, CLI, SDK, PRs, JSON, and Webhooks. Challenges include navigating Shadow DOM issues in web automation and balancing functionality with security. The developer has spent approximately $5,000 on tokens for the project.

Conclusion

Clubbot represents a significant step towards democratizing AI by providing an accessible, open-source framework for building personal agents. Its success hinges on continued community involvement and a commitment to user control and security. While challenges remain in scalability and model optimization, the project’s rapid growth and innovative approach position it as a key player in the evolving landscape of personal AI. The emphasis on repurposing older hardware and lowering the barriers to entry underscores a commitment to sustainability and inclusivity within the AI space.