Features
Self-hosted deployment on own infrastructure
Model-agnostic agent execution (Claude, GPT, Gemini, etc.)
Live per-step traces of tool calls, retrievals, reasoning
Replay and side-by-side comparison of agent runs
Audit-ready output with source citations
Real-time drift detection and alerts
Code Mode for up to 98% fewer tokens per run
Secure sandboxed execution for tool calls
Integration with MCP servers
Supports 50+ data types (emails, docs, spreadsheets, calls)
Open source and MIT-licensed
Air-gapped deployment support
Built on Agent-Verifiable Protocol (AVP)
Automatic model switching on provider downtime
On-prem and air-gapped deployment
Trace visibility for agent steps (prompts, retrievals, tool calls, outputs)
LLM-as-judge evaluation for relevance, toxicity, quality scoring
Dataset creation from traces for reproducible testing
Experiment management and regression benchmarking
Built-in Prompt IDE for iterative prompt optimization
Self-hosted deployment on local, Docker, Kubernetes
Phoenix Cloud managed hosting option
Vendor-agnostic support for any model/framework
Native OpenTelemetry integration
OpenInference specification for LLM telemetry
Human annotation and automated labeling
Ghost trajectories to simulate alternative agent paths
Eval-as-you-test for early quality feedback
One-click integration with LlamaIndex
