[opensourcedex]

coze-loop is a free, open source monitoring & observability project written in Go and released under Apache-2.0. It has 5,738 GitHub stars, 795 forks and 84 open issues, and was last pushed 2 hours ago. On this registry it ranks #83 of 191 tracked projects in Monitoring & Observability, with 5 head-to-head comparisons available.

What is coze-loop?

Coze Loop is a developer-oriented, open-source, platform-level solution for the full lifecycle management of AI agents — covering prompt development, debugging, evaluation, and observability — aimed at developers and teams who build and operate AI agents.

What it is

Coze Loop is an open-source edition of a commercial AI agent development and operation platform, released under the Apache-2.0 licence and written primarily in Go. It provides full lifecycle management capabilities across the agent development process: development, debugging, evaluation, and monitoring. The project is a developer-facing solution rather than an end-user agent product, and it is built around a shared core technology framework that developers can customize and extend according to business needs.

The concrete problem it solves is that agent development lacks a single place to write and version prompts, test them scientifically, and then watch what actually happens after deployment. Coze Loop replaces the fragmented combination of ad-hoc prompt files, one-off evaluation scripts, and separate tracing setups that teams stitch together themselves. It lives in the AI agent and LLMOps ecosystem, alongside the agent frameworks and tooling named in its own topic list such as LangChain and Eino, and it is tied to the Coze agent platform. Developers get a visual Playground for interactive prompt testing, systematic multi-dimensional evaluation, and end-to-end trace recording from user input through to AI output.

Key capabilities

  • Playground debugging and comparison, with real-time interactive prompt testing so developers can compare the output of different LLMs side by side.
  • Prompt version management covering writing, debugging, optimizing, and versioning of prompts end to end.
  • Evaluation tooling for managing evaluation sets, managing evaluators, and managing experiments.
  • Automated multi-dimensional evaluation of prompts and Coze agent output on dimensions such as accuracy, conciseness, and compliance.
  • SDK trace reporting plus trace data observation for the whole execution process.
  • Automatic capture of intermediate results and exceptions across key stages including prompt parsing, model invocation, and tool execution.
  • Model integration with OpenAI, Volcengine Ark, and other models, configured through release/deployment/docker-compose/conf/model_config.yaml.

Who uses it and how

  • Agent developers doing prompt engineering, who use the Playground for interactive iteration and comparison instead of editing prompt strings blind.
  • Teams that need reproducible quality checks, running evaluation sets and experiments to score prompts and Coze agents against accuracy, conciseness, and compliance criteria.
  • Operators of deployed agents, who rely on full execution traces to find where prompt parsing, model invocation, or tool execution broke down.
  • Engineering groups that need to extend the framework, since the open-source edition exposes the core framework for customization against business needs.
  • Community contributors and newcomers exploring agent practice, which the project explicitly targets with zero-barrier participation.

Getting started

The README documents Docker deployment via Docker Compose: clone https://github.com/coze-dev/coze-loop.git, enter the coze-loop directory, edit release/deployment/docker-compose/conf/model_config.yaml to set the api_key and model fields, then start the services, with Docker Engine installed and running first. The project points to its Quick Start wiki page for full installation and deployment details.

How it compares

Among the tools named in its own topics, Coze Loop positions itself as the lifecycle, evaluation, and observability layer rather than an agent orchestration framework like LangChain or Eino, and it integrates external model providers such as OpenAI and Volcengine Ark instead of supplying models itself. It is also the open-source edition of a commercial Coze Loop product, so its scope is deliberately the core foundational feature modules rather than the full commercial feature set.

When to use it — and when not to

A self-hoster must operate the Docker Compose stack and supply their own model credentials in model_config.yaml, whether from Volcengine Ark, OpenAI, or another supported provider, since the platform does not ship a model. Teams that only need simple prompt experimentation, or that already have mature tracing and evaluation pipelines wired to their own stack, should look elsewhere before adding another service to run. The README excerpt is also partially truncated and defers setup detail to an external wiki, so anyone without a working Docker environment and access to a supported model endpoint should expect extra legwork before the platform is useful.

project readme (upstream, from github) — read inline

Image

Coze Loop • Feature list • Quick start • Developer guide

License Go Version Ask DeepWiki

English | 中文

What is Coze Loop

Coze Loop is a developer-oriented, platform-level solution focused on the development and operation of AI agents. It addresses various challenges faced during the AI agent development process, providing full lifecycle management capabilities from development, debugging, evaluation, to monitoring.

Based on the commercial version, Coze Loop introduces an open-source edition that offers developers free access to core foundational feature modules. By sharing its core technology framework in an open-source model, developers can customize and extend according to business needs, facilitating community co-construction, sharing, and exchange, helping developers participate in AI agent exploration and practice with zero barriers.

What can Coze Loop do?

Coze Loop helps developers develop and operate AI Agent more efficiently by providing full lifecycle management capabilities. Whether it is prompt engineering, AI Agent evaluation, or monitoring and optimization after deployment, Coze Loop provides powerful tools and intelligent support, greatly simplifying the development process of AI Agents and enhancing their operational performance and stability.

  • Prompt development: The Prompt development module of Coze Loop provides developers with end-to-end support for writing, debugging, optimizing, and version management. Through a visual Playground, it enables real-time interactive testing of prompts, allowing developers to intuitively compare the output of different LLMs.
  • Evaluation: The Coze Loop evaluation module provides developers with systematic evaluation capabilities, enabling automated multi-dimensional testing of prompts and Coze agents' output, such as accuracy, conciseness, compliance, and more.
  • Observability: Coze Loop provides developers with observability for the entire execution process, fully recording every stage from user input to AI output, including key stages such as prompt parsing, model invocation, and tool execution, and automatically capturing intermediate results and exceptions.

Feature list

Feature Functional points
Prompt debugging Playground debugging and comparison
Prompt version management
Evaluation Manage evaluation sets
Management evaluator
Manage experiments
Observation SDK trace reporting
* Trace data observation
Model Support integration with OpenAI, Volcengine Ark, and other models

Quick Start

Refer to Quick Start to learn in detail how to install and deploy the latest version of Coze Loop.

Deployment method 1: Docker deployment (Docker Compose)

Please install and start Docker Engine before you start.

Procedure:

  1. Clone the source code. Run the following command to obtain the latest version of the Coze Loop source code.

    # Clone the code
git clone https://github.com/coze-dev/coze-loop.git

# Enter the coze-loop directory
cd coze-loop

  2. Configure a model.

    1. Enter the coze-loop directory.
    2. Edit the file release/deployment/docker-compose/conf/model_config.yaml.
    3. Modify the api_key and model fields. Take Volcengine Ark as an example:

  3. Start the service. Run the following commands to quickly deploy the open-source version of Coze Loop using Docker Compose.

    # Start the service (default: development mode)
# Run in the coze-loop/ directory
make compose-up

  4. Access the Coze Loop open-source version through your browser http://localhost:8082.

Deployment method 2: Kubernetes deployment using Helm Chart

  • The Kubernetes cluster has been prepared, the Nginx Ingress add-ons have been enabled, and the Kubectl and Helm tools have been installed.
  • To quickly try it out locally, you can deploy a Kubernetes cluster using Minikube. For detailed steps, refer to Quick Start.

Procedure:

  1. Run the following command to obtain the Helm Chart package.

    helm pull oci://docker.io/cozedev/coze-loop --version 1.0.0-helm
tar -zxvf coze-loop-1.0.0-helm.tgz && cd coze-loop && rm -f ../coze-loop-1.0.0-helm.tgz

  2. Configure a model. Go to the coze-loop directory and edit the release/deployment/helm-chart/umbrella/conf/model_config.yaml file. Configure the following fields, using Volcengine Ark as an example:

  3. Configure Ingress rules. Ingress is used to expose services to external networks. You need to configure the templates/ingress.yaml file in the project directory according to the actual cluster situation, manually modify parameters such as ingressClassName, and configure elements such as class, instance, host, and IP allocation.

  4. Deploy and start the service. Execute the following commands to quickly deploy the open-source version of Coze Loop using Helm.

    # Run in the coze-loop/ directory
make helm-up
# After the service deployment is complete, check the status of the cluster pods
make helm-pod
# Check the service startup logs. If both the app and nginx are running normally, the deployment is successful
make helm-logf-app
make helm-logf-nginx

  5. Access the Coze Loop open source edition via a browser. The access domain name and URL depend on the domain name and URL assigned to your cluster.

  6. Start customizing your Coze Loop project. Refer to the examples in the examples/ directory. Modify values.yaml to override the default settings. After making changes, rerun make helm-up for the changes to take effect.

[!WARNING] If you want to deploy Coze Loop in a public network environment, it is recommended to assess security risks before you begin, and take corresponding protection measures. Possible security risks include account registration functions, Coze Server listening address configurations, SSRF (Server - Side Request Forgery), and some horizontal privilege escalations in APIs. For more details, refer to Quickstart.

Use the Coze Loop open source version

  • Prompt development and debugging: Coze Loop provides a complete prompt development workflow.
  • Evaluation: The evaluation functionality of Coze Loop provides standard evaluation data management, an automated evaluation engine, and comprehensive statistics on experimental results.
  • Trace reporting and query: Coze Loop supports automatic reporting of traces from prompt debugging sessions created on the platform, enabling real-time

readme truncated — read the full docs on github

Frequently asked questions

Is coze-loop free to use?

coze-loop is open source under the Apache-2.0 licence. There is no licence fee and no seat count — you can self-host it or, where the project offers one, pay a vendor for a managed version instead.

What does coze-loop do?

Next-generation AI Agent Optimization Platform: Cozeloop addresses challenges in AI agent development by providing full-lifecycle management capabilities from d

What is coze-loop written in?

coze-loop is primarily written in Go. Its source is publicly available at https://github.com/coze-dev/coze-loop, and it has 5,738 GitHub stars.