Watchtower — Tracing, cost and PII masking for LLM apps

Traces, prompt governance, cost per tenant and PII masked at the source.

View on GitHub · PyPI

One record of everything the agent did.

Your first real trace in four steps.

  1. Start the dashboard: Two commands bring up the local UI.
  2. Copy the connection values: Paste them into your Python project.
  3. Configure once: Call wt.setup at application start.
  4. Send a trace: Run your app and watch it appear.

Details

Language
Python 3.11+
Dependencies
None required
Works with
OpenAI, Anthropic, LiteLLM, LangGraph
Sinks
Langfuse, Phoenix, LangSmith, your database
License
MIT

More from ShipIt