Cicaddy is a Python-based framework that enables agentic AI workflows to run directly inside existing CI/CD pipelines without requiring a dedicated agentic platform. It uses a one-shot execution model with multi-turn reasoning (ReAct pattern), supports multiple LLMs (Gemini, OpenAI, Claude), and integrates with MCP servers for external service access. Three agent types map to CI triggers: MR Agent for merge requests, Branch Agent for push events, and Task Agent for scheduled/manual runs. Real-world use cases include automated service health monitoring, DORA metrics reporting, and AI-powered code review. The approach leverages containerized CI environments for security isolation, uses DSPy YAML task definitions for structured prompts, and allows teams to share workflows as CI templates with a single include statement.
Table of contents
CI as an agentic runtimeOne-shot execution with multi-turn reasoningWhat is cicaddy?How MCP servers become your integration layerPipeline as workflow definitionSuccess story: GitLab AI Agent in productionMixing deterministic and AI processing in a pipelineStructured prompts for reproducible workflowsWhy this approach worksWrap upSort: