Problem solved

Terminal-based AI agents like Claude Code and Gemini CLI typically operate in isolation, each running in its own session without the ability to communicate or coordinate with others. This limits complex workflows where one agent's output could inform another's input or where multiple agents need to collaborate on a task. hcom solves this by providing a communication layer and process management that enables AI agents to message each other directly, watch each other's activities, and spawn new agent processes. It standardizes prompt passing and session handling across different tools, making it possible to build multi-agent systems entirely within the terminal environment.

Use cases

• Developers can run Claude Code and Gemini CLI simultaneously, with one agent reviewing and commenting on the other's code suggestions in real-time.
• Setting up a watchdog agent that monitors output logs from another agent and triggers corrective actions when errors are detected.
• Orchestrating a multi-step development pipeline where an analysis agent spawns a coding agent, then a testing agent, passing context via messages.
• Running experiments comparing outputs from different AI models by sending the same prompt to multiple agents and collecting responses.
• Building an agent mesh that persists across terminal sessions, with automatic orphan recovery ensuring reliability.
• Integrating hcom into CI/CD workflows to leverage multiple AI coding assistants for automated code generation and review.

Audience

• Developers
• ML Engineers

Pros

• Enables cross-agent communication between different AI CLI tools, reducing tool lock-in.
• Rust-based TUI provides a fast, terminal-native interface with low overhead.
• Supports headless mode for scripted or automated workflows.
• Plugin architecture allows integration of new AI tools (e.g., Claude Code plugin).
• Orphan recovery mechanism improves reliability for long-running agent sessions.

Cons

• Requires users to install and configure multiple separate AI CLI tools beforehand.
• Documentation may be limited, relying heavily on the GitHub README and commit logs.
• Early-stage project (version 0.6.x) with potential breaking changes.