coze-loop

Coze Loop is an open-source, developer-oriented platform-level solution for building and operating AI agents. Based on a commercial edition, the open-source edition exposes core foundational modules so developers can customize and extend them for their own needs. Core capabilities described in the README: - Prompt development: a visual Playground for writing, debugging, optimizing and version-managing prompts, with real-time interactive testing and comparison of outputs across different LLMs. - Evaluation: systematic, automated multi-dimensional testing of prompts and agent outputs (for example accuracy, conciseness, compliance), plus management of evaluation sets, evaluators and experiments. - Observability: full execution tracing from user input to AI output, recording prompt parsing, model invocation and tool execution, and capturing intermediate results and exceptions. Traces can be reported via SDK. - Model integration: support for OpenAI, Volcengine Ark and other models, configured through the Eino framework. Deployment options: - Docker Compose: clone the repository, edit release/deployment/docker-compose/conf/model_config.yaml to set api_key and model, then run make compose-up and open http://localhost:8082. - Kubernetes with Helm: pull the OCI chart, configure model_config.yaml, adjust Ingress rules, then use make helm-up, make helm-pod and log commands to verify the deployment. The README warns that public-network deployments should be security-assessed, citing account registration, Coze Server listening address configuration, SSRF and API privilege escalation risks. The project is licensed under Apache 2.0 and welcomes community contributions through GitHub issues and pull requests, with discussion channels on Lark, Discord and Telegram.