What is tellmphone?

TeLLMphone is a tool that lets different AI coding assistants talk to each other on the same computer, instead of you manually copying answers between them. If you use one AI assistant like Claude Code but sometimes want a second opinion from another one like Codex, Gemini, or Grok, this project sets up a local switchboard so your first assistant can just call the second one directly, ask a question about your project, and relay the answer back to you. It runs as what is called an MCP server, a kind of local connector that plugs into the command line tools for these AI assistants. Once installed, it registers itself with whichever assistant CLIs are found on your machine. From then on, you can simply ask your assistant, in plain language, to call another assistant and get its opinion, optionally telling it to use a specific personality, like a strict reviewer or a skeptical devil's advocate. The called assistant runs its own turn in the background and its answer lands back with the assistant that made the call, and the conversation can continue with follow up messages. Under the hood there is no constantly running background service. Each assistant's copy of TeLLMphone talks over a simple local channel, and calls are handled by starting one background run of the other assistant's command line tool, then returning its answer once finished. Called assistants run in a safe, read only mode by default unless a project folder is specifically allowed to have changes written to it, and there is a limit on how many times assistants can call each other in a chain, to prevent them looping forever. The project ships with thirteen ready made personalities, such as a grumpy reviewer, a security auditor, and a rubber duck style thinking partner, and you can add your own as simple text files. All call records and conversation transcripts are stored as plain files on your own machine, which you can delete at any time. This requires macOS or Linux, Python, and the command line tools for whichever AI assistants you want connected, already installed and logged in. It is released under the Apache 2.0 license.