ctgbot

ctgbot serves as a substrate for AI agent ecosystems where specializations and relationships emerge through use rather than predefined roles. It focuses on explicit trust boundaries and isolated environments to allow agents to operate as long-term entities rather than ephemeral task runners. Key capabilities include: - **Isolated Runtimes**: Each agent conversation runs in its own Docker container, allowing agents to install packages, run tests, and build tools without affecting the host. - **Persistent Identities**: Agents maintain conversation history and accumulate context across sessions, supported by semantic long-term memory that is searchable across threads. - **Hostbridge**: A typed, authenticated command bridge that allows agents to interact with the host system, send messages to other agent threads, and execute restricted host command aliases. - **Component Model**: A modular architecture where agents (e.g., Claude, Codex), sources (e.g., Gmail), and filters can be mixed and matched per chat. - **Explicit Trust Model**: Implements inbound filters and allowlists; unknown communication channels are dropped until explicitly bound by an operator. - **Inter-Agent Messaging**: Agents can coordinate by sending messages, files, and structured commands to other agent threads via the hostbridge. The platform supports various integrations including Telegram for user interaction and Gmail for inbound/outbound communication.