TraceAgently is a real-time observability and monitoring tool for AI agents. It shows agent activity as it happens and is designed to help control runaway agents before they drain budget. The product describes itself as working with agents in real time, with live traces, an automatic kill-switch, and loop detection.
Its tracing view captures thoughts, tool calls, tool results, errors, timestamps, token counts, and cost per step. The feature set also includes trace comparison, which presents two runs side by side and compares cost, tokens, duration, and event count. Users can mark a run as Golden and compare future traces against it. Additional features mentioned include cost tracking, email alerts when a kill-switch fires or a loop is detected, error pattern detection that groups and ranks recurring errors, and Magic Fix, which sends a trace to Claude for diagnosis and a suggested fix.
TraceAgently is described as framework-agnostic and works with LangChain, CrewAI, raw OpenAI calls, or a custom loop. It provides SDKs for Python and TypeScript, with installation shown through pip and npm, and the examples use API keys and tracing calls in code. The site also says instrumenting takes 2 lines of code, and the quick start says setup can be done in under 2 minutes.
Pricing includes a free plan with up to 1,000 traces per month, 7-day retention, and one agent, with no credit card required. A Pro plan is listed at $49 per month, with 1,000,000 traces per month, 90-day retention, unlimited agents, and features including kill-switch, loop detection, Golden traces, trace comparison, Magic Fix, email alerts, and error pattern detection. The product is presented as a real-time observability tool for AI agents.
traceagently is a LLM eval & observability project. It focuses on providing real-time observability and debugging for AI agent workflows. It is built as an open-source project for AI developers. The project is open source (MIT). traceagently is available on the web and the command line.
traceagently first shipped in 2026. Among its 5 catalogued features are real-time tracing, agent observability, and error tracking.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do