What we're building

Auto-instrumentation for multi-agent frameworks - trace delegation, handoffs, and per-agent metrics.

Automated prompt improvement suggestions based on evaluation results and production traces.

Sub-100ms inline guardrails for PII detection, topic enforcement, and hallucination blocking.

One-command self-hosted deployment with Docker Compose and Helm charts for air-gapped environments.

End-to-end tracing for browser-based voice agents - ICE metrics, audio pipeline, and STT/TTS latency.

Version-controlled evaluation datasets with diff views, annotations, and CI/CD integration.

Trace tool calls and context flows in MCP-enabled agents.

Native TypeScript SDK with the same auto-instrumentation capabilities as the Python SDK.

Built-in prompt experimentation with traffic splitting, statistical significance, and auto-rollback.

Push hallucination and latency alerts directly to Slack channels and PagerDuty incidents.

Define and register your own evaluation metrics with a simple Python decorator API.

Per-agent, per-model token cost tracking with budget alerts and optimization recommendations.

Full distributed tracing for LLM calls, tool use, and agent handoffs using OpenTelemetry.

Automated agent-to-agent voice testing - simulate 1000s of call scenarios without human testers.

Zero-code tracing for LangChain chains and LlamaIndex queries with full context propagation.

Built-in factuality, groundedness, and faithfulness scoring for RAG and agent outputs.

One-line instrumentation for Vapi and Retell voice agent platforms.