Observability
Basalt v1 introduces a unified, OpenTelemetry-based observability system for AI applications. It replaces the v0 tracing APIs with a single, coherent model built aroundstart_observe and observe.
With Observability, you can:
- Trace complete AI workflows (prompts, RAG, tools, evaluators) as a single trace
- Attach user and organization identity to requests and have it flow automatically
- Track latency, token usage, and errors for LLM calls and tools
- Enable auto-instrumentation for LLM SDKs and other libraries with minimal code
This section explains how Basalt Observability works conceptually and how it differs from v0.
For detailed Python usage, see the Python pages in this section. TypeScript v1 docs are not available yet (use the v0 archive).
For detailed Python usage, see the Python pages in this section. TypeScript v1 docs are not available yet (use the v0 archive).
What’s new in v1
- Unified
observe/start_observeAPI for tracing, logging, and context - Full OpenTelemetry compatibility and context propagation
- Auto-instrumentation for LLMs, vector DBs, and frameworks
- Rich identity, experiment, and evaluator tracking
- Async and sync support with the same decorator and context manager APIs
Migration: v1 is not backward compatible with v0. See the Python Observability docs for migration guidance and side-by-side examples.