openinference

Multi-language agent runtime and library for execution scope management, lifecycle events, and middleware on tool and LLM calls.

OTel Plugin for Hermes Agent

Open your AI agents' black box. The waste is between your agents, not inside them. Four deterministic detectors over one trace file, priced from real usage fields. Hosted at boxdawn.com.

Rust-first, local-first observability & debugging for LLM agents: capture, inspect, replay, diff, and export traces (OpenTelemetry-compatible).

Find where AI apps break first — and block regressions before they ship.

Agent-native OTel router. One-line auto-instrument for Anthropic / OpenAI / OpenRouter / Vercel AI. Counterfactual replay, reversible PII masking, `scry` CLI for agents to debug themselves. Composes with Braintrust / Phoenix / Langfuse — don't replace.

TracePilot: Gemini CLI fork with Phoenix/OpenInference tracing, MCP self-introspection, safety gates, redaction, evals, and a verified broken-repo repair loop

BUFI Minions — self-improving Gemini 3 agents with Arize Phoenix observability (OpenInference tracing, Phoenix MCP self-introspection, LLM-as-judge evals). Devpost 'Agents for Real-World Challenges' Arize track.

A truth-first visual observatory for understanding, replaying, comparing, and monitoring AI agent runs.

Signed, tamper-evident provenance records for AI agent runs

Next.js demo: Scalekit AgentKit tool executions and OpenAI LLM calls in one Arize AX trace — public-API wrapper, not a prototype patch.

🛡️ Catch unsafe AI agent actions before they execute — explainable risk scoring, injection detection, and real-time containment powered by SigNoz & OpenTelemetry.

Passive OpenInference/OTLP tracing adapter for Deep Agents Code Hooks v2

🛡️ AEGIS is an autonomous meta-agent that watches, detects loops in, and heals other AI agents in real-time.

Multi-harness coding-agent tracing with durable SQLite request and session analytics

Retry Safety Analyzer: deterministic HIGH/MEDIUM/SAFE when a side-effecting tool may have been retried. Offline import — never executes the trace. HIGH is risk, not proof. External validation open.

Catch the step where an AI agent contradicts a decision it already made. Adapters and a CLI for the Fathom committed-state read over LangGraph, CrewAI, Letta, OpenInference, DBOS, and coding-agent traces.

Production tracing layer for Claude Agent SDK → Langfuse: named traces, session/user injection, usage & cost normalization, redaction, and full-chain CLI OpenTelemetry.

Deliberately breakable order-status agent for learning MCP, OpenTelemetry, observability, evaluation, and k3s.