Phoenix integrates with the leading AI frameworks, LLM providers, and tools to provide seamless observability, evaluation, and debugging for your AI applications. Whether you’re building with Python, TypeScript, or Java, Phoenix has you covered.
Don’t see an integration you need? We’d love to hear from you!
Integration Types#
Phoenix offers several types of integrations to support your AI development workflow:
Integrate Phoenix with AI coding assistants like Claude Code and Cursor for debugging and analysis.
Automatically capture traces from your AI applications built with popular frameworks and LLM providers.
Use any LLM provider to power Phoenix’s evaluation capabilities for scoring and classifying your traces.
Record results from evaluation frameworks and libraries such as Harbor, Ragas, Cleanlab, and MLflow.
Convert traces from other instrumentation libraries to the OpenInference format.
Run code evaluators in hosted sandbox providers with kernel-level isolation.
Developer Tools#
Integrate Phoenix with AI coding assistants to debug and analyze your LLM applications directly from your development environment.
Coding Agents#
Trace your sessions with a coding agent — turns, tool calls, and token costs — in Phoenix, no application code changes required. Install the coding-harness-tracing toolkit and pick your agent:
Tracing Integrations#
Phoenix captures detailed traces from your AI applications, giving you visibility into every step of your LLM pipeline.
By Language#
LLM Providers#
Phoenix provides native tracing support for all major LLM providers:
Platforms#
Integrate Phoenix with AI development platforms and infrastructure:
Vector Databases#
Trace retrieval from vector databases to see which stage of a search to fix:
Eval Model Integrations#
Phoenix’s evaluation library (phoenix-evals) can use any LLM provider to power evaluations. These models score, classify, and analyze your traces.
Evaluation Integrations#
Connect external evaluation frameworks and libraries to Phoenix:
Span Processors#
Normalize and convert data from other instrumentation libraries by adding span processors that unify traces to the OpenInference format:
Sandboxes#
Run Phoenix code evaluators in hosted sandbox providers for kernel-level isolation, runtime dependency installation, and opt-in outbound network access. Configure providers from Settings → Sandboxes.