learningCatHD/telos-sdk — reverse-engineered prompt

Build me a Python tool called TELOS that helps people spend less on AI coding agents without changing how they work. I want it to run locally as a small gateway between tools like Claude Code, Codex, openclaw, or Hermes and the model provider, then make repeated context cheaper by using the model cache instead of sending the same full prompt every turn.

Make it easy for a normal developer to install, run telos init, have it detect supported agents, start the local gateway, and then open telos dashboard to see how much money and input tokens were saved. The dashboard should be an offline HTML page with clear savings by session, model, and tool.

Please keep privacy as a core promise. Do not store prompt or response text, only usage numbers like token counts and cost estimates. Add clear docs, a quickstart, examples, tests, and a simple CLI. Look up current docs online if you need to.