Docs

Torch Log Service

AgentLoop

AgentLoop overview

This topic describes the features of the TLS AgentLoop platform.

Product overview

AgentLoop is a platform for monitoring Agent applications based on TLS, including runtime tracking, issue archiving, offline evaluation, experiment comparison, and quality analysis. The platform supports full observability of runtime process, replay of complex workflows, issue archiving for online scenarios, and evaluation on version optimization.

With the TLS AgentLoop platform, you can access Agent applications with one click. What's more, the system provides built-in evaluators for multiple scenarios. With Coze Loop as the evaluation engine, you can quickly create evaluation sets and experiments, and use experimental data analysis reports to verify and optimize evaluation targets.

Billing overview

AgentLoop features on TLS are free of charge, but costs will be generated in terms of log storage, data transformation, and log shipping.

Storage costs for accessing log applications are the same as standard TLS log topics.

If the data backflow target is a TLS log topic, costs are the same as for a standard log topic.

If the data backflow target is set to ship logs to AgentLoop evaluation sets and TOS, costs include TLS log shipping fees and TOS storage fees. For details, see Overview of shipping logs to TOS.

There is currently no charge for data review features. We will release related notes on billing time and billing rules in the upcoming product announcement. For details, please stay tuned. However, charges related to models used for evaluation and experiments are collected by Volcano Ark. For more information, see model pricing.

Core features

The TLS AgentLoop platform core features include:

Feature highlights	Notice
Data integration	Through the LLM Application Marketplace of TLS, you can integrate Agent monitoring data into TLS with one click. For details, see Data integration. Supports LLM application monitoring, VeADK monitoring, OpenClaw monitoring, HermesAgent monitoring, and more. Create separate log topics for different types of log applications to achieve physical data isolation and categorized storage.
Data Monitoring	After integrating log applications, TLS provides a professional data visualization dashboard. For details, see Overview of log applications. By configuring data source mapping, axes, legends, and other settings in the application template, users can generate visual charts by simply linking their own log fields.
Data backflow	Import observability data obtained through TLS into the AgentLoop evaluation set by creating data backflow tasks. For details, see Data backflow overview. This feature monitors key node data from actual user requests and their processing, helping identify unknown issues. Supports on-demand backflow of online data, including end-to-end input and output, LLM input and output, and data used to supplement evaluation or training sets for ongoing AI Agent optimization.
Data evaluation	Import data from AgentLoop evaluation sets via data backflow tasks for evaluation by creating experiments. For details, see Evaluation overview. TLS emmplys Coze Loop as the evaluation engine, which offers standardized data management, automated evaluation, and comprehensive experiment result statistics to help developers optimize AI Agent effectiveness, performance, and cost. Evaluation set: a group of data used to assess evaluation targets, typically including input data and expected output results. Evaluator: Analyze the output of the evaluation object from multiple dimensions using predefined rules and generate measurable metrics and attribution conclusions. Experiment: Combine evaluation datasets, evaluation objects, and several evaluator triplets to execute evaluation actions and obtain experimental results.

Steps

The TLS AgentLoop workflow is as follows:

Last updated: 2026.07.30 18:21:21