TricorderAgent — public web build

Bring your own model

This is the TricorderAgent interface running as a plain web page — no account, no backend, no key of mine. It talks straight to a model server you run, and the conversation never leaves your browser.

Two ways to use this page. Point it at your own LM Studio, llama.cpp or Ollama server and it becomes a working agent with in-browser tools. Without a server, run the recorded walkthrough to see how a real session looks — it is a replay, and every line of it is labelled as one.

The agent speaks one dialect: an OpenAI-compatible /v1/chat/completions endpoint. Anything that serves it works — on your machine, on your LAN, or a hosted provider. Nothing you type here is sent anywhere except to the host you name; it is stored in this browser only.

Preset

Model host URL Base URL without /v1. LM Studio uses :1234, llama.cpp :8080, Ollama :11434. API key — optional Sent as a bearer token to the host above and to nothing else. Kept in this browser's local storage — clear it with “Forget everything” below.

Model Built and tuned against Qwen3.8-27B, but any tool-calling model works. Context window Only used for the CTX meter and auto-trimming. Set it if the meter looks wrong.

Browser reality check