Introduction
TokenSense is an open-source Python framework for LLM cost observability and smart model routing. It wraps your existing LLM clients with one line of code and captures metadata — tokens, cost, latency — after every call. It is not:- A proxy — your API keys and prompts never go through our servers
- A SaaS platform — no account, no sign-up, nothing to install except the package
- A dashboard — it outputs to wherever you point it
- A Python package that runs inside your process
- A transparent wrapper around OpenAI, Anthropic, and Groq clients
- A routing layer that is context-budget-aware, rule-based, and auditable
- Private by default — prompts are never captured unless you explicitly opt in