wtf is lanes?

Lanes is a small local tool that lets you hold several conversations with different AI models at the same time, side by side, in one browser window. It shows between one and six independent chat lanes, four by default, and each lane displays live performance numbers above the conversation, such as how quickly the first word appears, how many tokens per second the model is generating, total output tokens, and total response time. The project is a fork of an existing benchmark tool by another developer, which originally measured coding agent latency. Lanes keeps that tool's core building blocks, a tiny server using only Python's standard library, real streaming performance numbers, and the ability to cancel an in progress response, but changes its purpose from one off benchmark runs into ongoing, independent multi turn chats. Each lane can talk to any model listed in a local configuration file that follows the OpenAI compatible API format. You can type a message into one lane, or use a Send All button to broadcast the same prompt to every visible lane at once and compare how different models respond. A shared Crew Room panel lets multiple lanes post short results into a common local chat log, so several models can share a running discussion thread. A Kill All button, or the Escape key, stops every busy lane immediately. Lane settings and recent conversation history are saved in the browser's local storage so they persist between sessions. Lanes relies on a separate tool called XAV to actually load and unload AI models and manage how much memory each one uses, meaning Lanes itself focuses purely on the chat interface while XAV handles model residency and memory rules behind the scenes. It runs entirely on your own machine and binds to localhost by default. The project is released under the MIT license.