bcurts/agentchattr

What it solves

AgentChattr eliminates the funerals of copy-pasting prompts between multiple AI coding agents and terminal windows. It provides a shared, real-time chat interface where humans and multiple AI agents (like Claude Code, Gemini CLI, and GitHub Copilot CLI) can coordinate and communicate in a single room.

How it works

The system acts as a local chat server that wraps around existing CLI-based AI agents. When a user or another agent @mentions a specific agent in the web UI, the server automatically injects a prompt into that agent's terminal, instructing it to read the current channel's history via MCP (Model Context Protocol) and respond. This creates a hands-free loop where agents can wake each other up and collaborate autonomously.

Who it’s for

Developers who use multiple AI coding agents simultaneously and want a centralized hub for coordination, task tracking, and multi-agent orchestration without manual terminal management.

Highlights

• Multi-Agent Coordination: Supports a wide array of agents (Claude, Codex, Gemini, Copilot, Kimi, Qwen, Kilo, CodeBuddy, MiniMax) and any MCP-compatible agent.
• Agent-to-Agent Communication: Agents can @mention each other to trigger responses, with a loop guard to prevent runaway conversations.
• Structured Workflows: Includes "Jobs" for tracking bounded tasks and "Sessions" for orchestrated multi-agent phases (e.g., code reviews or debates).
• Behavior Steering: Allows assigning persistent roles (e.g., Planner, Reviewer) and defining active rules to guide agent behavior.
• Interactive UI: Features inline decision cards for yes/no or multiple-choice questions, real-time activity indicators based on terminal buffer hashing, and channel-based organization.
• Project Management: Supports pinned messages, image sharing, and full project history export/import.
• Crosstalk Tools: Includes channel summaries to help agents catch up quickly and scheduled messages for recurring prompts.