AgentX, a new experimental project by developer Anis Marrouchi, has landed on Hacker News with a promise to let users run Claude Code, Codex, or any LLM directly from their existing chat channels and GitHub workflows. The tool acts as a routing layer, connecting platforms like Telegram, WhatsApp, GitLab, and GitHub to AI agents hosted on your own machine. While the concept is compelling for those tired of switching between browser tabs and terminals, the project is explicitly labeled as 'usable, experimental, not yet stable,' warning early adopters to expect rough edges in the Settings experience and configuration flows.

The Architecture of Self-Hosted Agents

The core value proposition of AgentX is decentralization. Unlike SaaS-heavy agent frameworks, AgentX requires Node.js 22.19 or newer and runs locally. It supports a mesh architecture where agents can be paired across different machines using agentx connect mesh invite and agentx connect mesh join. This allows for distributed workflows, such as a 'CX' agent on one node delegating a build investigation to a 'builder' agent on another machine when CI fails on GitLab. The system provides a browser-based dashboard with six top-level tabs: Live, Operations, Monitor, Activity, Workflows, and Settings, giving users real-time visibility into task execution.

Integration and Current Limitations

AgentX currently supports Telegram and WhatsApp via pairing flows, while GitLab, GitHub, and generic webhooks handle event-driven tasks. Notably, Slack and Discord are not supported as live channel adapters in this build. The tool integrates with API providers and local CLIs for Claude Code, Codex, and OpenCode. However, critical features like Role-Based Access Control (RBAC) are still missing. The developers have stated that RBAC at both mesh and agent levels is planned but not available, which limits its immediate utility for multi-user teams who need strict delegation and security boundaries.

Key Takeaways

  • AgentX is MIT-licensed and self-hosted, requiring no external accounts for the core demo.
  • It supports routing from Telegram, WhatsApp, GitHub, GitLab, and webhooks to local LLMs.
  • The project is experimental; users should expect instability and incomplete configuration flows.
  • RBAC is not yet implemented, restricting multi-user team deployment capabilities.
  • Installation is possible via Docker Compose or direct Node.js/pnpm builds.

The Bottom Line

AgentX is a fascinating glimpse into the future of decentralized AI orchestration, but it feels like a power-user prototype right now. If you are a developer who wants to hack on your own agent mesh and doesn't mind debugging config files, give it a spin. Otherwise, wait for the RBAC and stability patches before trusting it with production workflows.