AI/LLM-Powered Pipeline Analysis

Pipelines-as-Code can analyze your CI/CD pipeline runs using Large Language Models (LLMs). An LLM is an AI service (such as OpenAI or Google Gemini) that reads pipeline logs and failure data, then posts human-readable analysis directly on your pull requests.

Use this feature when you want automated root-cause analysis of pipeline failures without manually reading through logs.

Overview

LLM-powered analysis lets you:

• Analyze failed pipelines automatically and get root-cause analysis
• Generate actionable recommendations for fixing issues
• Post insights as PR comments so your team sees them immediately
• Configure custom analysis scenarios using different prompts and triggers

Supported LLM Providers

Pipelines-as-Code supports two LLM providers:

• OpenAI – Default model: gpt-5-mini
• Google Gemini – Default model: gemini-2.5-flash-lite

You can specify any model your chosen provider supports. See Model Selection for guidance on choosing the right model.

Configuration

To enable LLM-powered analysis, configure the spec.settings.ai section in your Repository CR:

apiVersion: pipelinesascode.tekton.dev/v1alpha1
kind: Repository
metadata:
  name: my-repo
spec:
  url: "https://github.com/org/repo"
  settings:
    ai:
      enabled: true
      provider: "openai"
      timeout_seconds: 30
      max_tokens: 1000
      secret_ref:
        name: "openai-api-key"
        key: "token"
      roles:
        - name: "failure-analysis"
          model: "gpt-5-mini"  # Optional: specify model (uses provider default if omitted)
          prompt: |
            You are a DevOps expert. Analyze this failed pipeline and:
            1. Identify the root cause
            2. Suggest specific fixes
            3. Recommend preventive measures
          on_cel: 'body.pipelineRun.status.conditions[0].reason == "Failed"'
          context_items:
            error_content: true
            container_logs:
              enabled: true
              max_lines: 100
          output: "pr-comment"

Configuration Fields

The following tables describe every field available in the spec.settings.ai configuration block.

Top-Level Settings

Analysis Roles

Each role defines a specific analysis scenario. You can configure multiple roles to handle different types of pipeline events (for example, one role for security failures and another for general build failures).

Context Items

Context items control what information Pipelines-as-Code sends to the LLM provider. Choose carefully, because more context means higher token usage and cost.

Commit Fields

When you set commit_content: true, Pipelines-as-Code includes the following fields in the data sent to the LLM provider:

Privacy and Security Notes:

• Pipelines-as-Code intentionally excludes email addresses from the commit context to protect personally identifiable information (PII) when sending data to external LLM providers.
• Fields appear only if your Git provider makes them available. Some providers supply limited information (for example, Bitbucket Cloud provides only the author name).
• Author and committer may be the same person or different (for example, when using git commit --amend or rebasing).