llm-ci-runner 1.4.1

A simple CI/CD utility for running LLM tasks with Semantic Kernel

AI-First Toolkit: LLM-Powered Automation

🚀 The Future of DevOps is AI-First
This toolkit represents a step toward AI-First DevOps - where intelligent automation handles the entire development lifecycle. Built for teams ready to embrace the exponential productivity gains of AI-powered development. Please read the blog post for more details on the motivation.

TLDR: What This Tool Does

Purpose: Zero-friction LLM integration for pipelines with 100% guaranteed schema compliance. This is your foundation for AI-first integration practices.

Perfect For:

• 🤖 AI-Generated Code Reviews: Automated PR analysis with structured findings
• 📝 Intelligent Documentation: Generate changelogs, release notes, and docs automatically
• 🔍 Security Analysis: AI-powered vulnerability detection with structured reports
• 🎯 Quality Gates: Enforce standards through AI-driven validation
• 🚀 Autonomous Development: Enable AI agents to make decisions in your pipelines
• 🎯 JIRA Ticket Updates: Update JIRA tickets based on LLM output
• 🔗 Unlimited Integration Possibilities: Chain it multiple times and use as glue code in your tool stack

---

Simple structured output example

# Install and use immediately
pip install llm-ci-runner
llm-ci-runner --input-file examples/02-devops/pr-description/input.json --schema-file examples/02-devops/pr-description/schema.json

The AI-First Development Revolution

This toolkit embodies the principles outlined in Building AI-First DevOps:

Traditional DevOps	AI-First DevOps (This Tool)
Manual code reviews	🤖 AI-powered reviews with structured findings
Human-written documentation	📝 AI-generated docs with guaranteed consistency
Reactive security scanning	🔍 Proactive AI security analysis
Manual quality gates	🎯 AI-driven validation with schema enforcement
Linear productivity	📈 Exponential gains through intelligent automation

Features

• 🎯 100% Schema Enforcement: Your pipeline never gets invalid data. Token-level schema enforcement with guaranteed compliance
• 🔄 Resilient execution: Retries with exponential back-off and jitter plus a clear exception hierarchy keep transient cloud faults from breaking your CI.
• 🚀 Zero-Friction CLI: Single script, minimal configuration for pipeline integration and automation
• 🔐 Enterprise Security: Azure RBAC via DefaultAzureCredential with fallback to API Key
• 📦 CI-friendly CLI: Stateless command that reads JSON/YAML, writes JSON/YAML, and exits with proper codes
• 🎨 Beautiful Logging: Rich console output with timestamps and colors
• 📁 File-based I/O: CI/CD friendly with JSON/YAML input/output
• 📋 Template-Driven Workflows: Handlebars and Jinja2 templates with YAML variables for dynamic prompt generation
• 📄 YAML Support: Use YAML for schemas, input files, and output files - more readable than JSON
• 🔧 Simple & Extensible: Easy to understand and modify for your specific needs
• 🤖 Semantic Kernel foundation: async, service-oriented design ready for skills, memories, orchestration, and future model upgrades
• 📚 Documentation: Comprehensive documentation for all features and usage examples. Use your semantic kernel skills to extend the functionality.
• 🧑‍⚖️ Acceptance Tests: pytest framework with the LLM-as-Judge pattern for quality gates. Test your scripts before you run them in production.
• 💰 Coming soon: token usage and cost estimation appended to each result for budgeting and optimisation

🚀 The Only Enterprise AI DevOps Tool That Delivers RBAC Security, Robustness and Simplicity

LLM-CI-Runner stands alone in the market as the only tool combining 100% schema enforcement, enterprise RBAC authentication, and robust Semantic Kernel integration with templates in a single CLI solution. No other tool delivers all three critical enterprise requirements together.

Installation

pip install llm-ci-runner

That's it! No complex setup, no dependency management - just install and use. Perfect for CI/CD pipelines and local development.

Quick Start

1. Install from PyPI

pip install llm-ci-runner

2. Set Environment Variables

Azure OpenAI (Priority 1):

export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com/"
export AZURE_OPENAI_MODEL="gpt-4.1-nano"  # or any other GPT deployment name
export AZURE_OPENAI_API_VERSION="2024-12-01-preview"  # Optional

OpenAI (Fallback):

export OPENAI_API_KEY="your-very-secret-api-key"
export OPENAI_CHAT_MODEL_ID="gpt-4.1-nano"  # or any OpenAI model
export OPENAI_ORG_ID="org-your-org-id"  # Optional

Authentication Options:

• Azure RBAC (Recommended): Uses DefaultAzureCredential for Azure RBAC authentication - no API key needed! See Microsoft Docs for setup.
• Azure API Key: Set AZURE_OPENAI_API_KEY environment variable if not using RBAC.
• OpenAI API Key: Required for OpenAI fallback when Azure is not configured.

Priority: Azure OpenAI takes priority when both Azure and OpenAI environment variables are present.

3a. Basic Usage

# Simple chat example
llm-ci-runner --input-file examples/01-basic/simple-chat/input.json

# With structured output schema
llm-ci-runner \
  --input-file examples/01-basic/sentiment-analysis/input.json \
  --schema-file examples/01-basic/sentiment-analysis/schema.json

# Custom output file
llm-ci-runner \
  --input-file examples/02-devops/pr-description/input.json \
  --schema-file examples/02-devops/pr-description/schema.json \
  --output-file pr-analysis.json

# YAML input files (alternative to JSON)
llm-ci-runner \
  --input-file config.yaml \
  --schema-file schema.yaml \
  --output-file result.yaml

3b. Template-Based Workflows

Dynamic prompt generation with YAML, Handlebars or Jinja2 templates:

# Handlebars template example
llm-ci-runner \
  --template-file examples/05-templates/handlebars-template/template.hbs \
  --template-vars examples/05-templates/handlebars-template/template-vars.yaml \
  --schema-file examples/05-templates/handlebars-template/schema.yaml \
  --output-file handlebars-result.yaml
  
# Or using Jinja2 templates
llm-ci-runner \
  --template-file examples/05-templates/jinja2-template/template.j2 \
  --template-vars examples/05-templates/jinja2-template/template-vars.yaml \
  --schema-file examples/05-templates/jinja2-template/schema.yaml \
  --output-file jinja2-result.yaml

For more examples see the examples directory.

Benefits of Template Approach:

• 🎯 Reusable Templates: Create once, use across multiple scenarios
• 📝 YAML Configuration: More readable than JSON for complex setups
• 🔄 Dynamic Content: Variables and conditional rendering
• 🚀 CI/CD Ready: Perfect for parameterized pipeline workflows

4. Development Setup (Optional)

For contributors or advanced users who want to modify the source:

# Install UV if you haven't already
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and install for development
git clone https://github.com/Nantero1/ai-first-devops-toolkit.git
cd ai-first-devops-toolkit
uv sync

# Run from source
uv run llm-ci-runner --input-file examples/01-basic/simple-chat/input.json

Real-World Examples

You can explore the examples directory for a complete collection of self-contained examples organized by category.

For comprehensive real-world CI/CD scenarios, see examples/uv-usage-example.md.

The AI-First Transformation: Why Unstructured → Structured Matters

LLMs excel at extracting meaning from messy text, logs, documents, and mixed-format data, then emitting schema-compliant JSON/YAML that downstream systems can trust. This unlocks:

• 🔄 Straight-Through Processing: Structured payloads feed BI dashboards, RPA robots, and CI/CD gates without human parsing
• 🎯 Context-Aware Decisions: LLMs fuse domain knowledge with live telemetry to prioritize incidents, forecast demand, and spot security drift
• 📋 Auditable Compliance: Formal outputs make it easy to track decisions for regulators and ISO/NIST audits
• ⚡ Rapid Workflow Automation: Enable automation across customer service, supply-chain planning, HR case handling, and security triage
• 🔗 Safe Pipeline Composition: Structured contracts let AI-first pipelines remain observable and composable while capitalizing on unstructured enterprise data

100 Diverse AI Automation Use Cases

DevOps & Engineering 🔧

1. 🤖 AI-generated PR review – automated pull request analysis with structured review findings
2. 📝 Release note composer – map commits to semantic-version bump rules and structured changelogs
3. 🔍 Vulnerability scanner – map code vulnerabilities to OWASP standards with actionable remediation
4. ☸️ Kubernetes manifest optimizer – produce risk-scored diffs and security hardening recommendations
5. 📊 Log anomaly triager – convert system logs into OTEL-formatted events for SIEM ingestion
6. 💰 Cloud cost explainer – output tagged spend by team in FinOps schema for budget optimization
7. 🔄 API diff analyzer – produce backward-compatibility scorecards from specification changes
8. 🛡️ IaC drift detector – turn Terraform plans into CVE-linked security findings
9. 📋 Dependency license auditor – emit SPDX-compatible reports for compliance tracking
10. 🎯 SLA breach summarizer – file structured JIRA tickets with SMART action items

Governance, Risk & Compliance 🏛️ 11. 📊 Regulatory delta analyzer – emit change-impact matrices from new compliance requirements 12. 🌱 ESG report synthesizer – map CSR prose to GRI indicators and sustainability metrics 13. 📋 SOX-404 narrative converter – transform controls descriptions into testable audit checklists
14. 🏦 Basel III stress-test interpreter – output capital risk buckets from regulatory scenarios 15. 🕵️ AML SAR formatter – convert investigator notes into Suspicious Activity Report structures 16. 🔒 Privacy policy parser – generate GDPR data-processing-activity logs from legal text 17. 🔍 Internal audit evidence linker – export control traceability graphs for compliance tracking 18. 📊 Carbon emission disclosure normalizer – structure sustainability data into XBRL taxonomy 19. ⚖️ Regulatory update tracker – generate structured compliance action items from guideline changes 20. 🛡️ Safety inspection checker – transform narratives into OSHA citation checklists

Financial Services 🏦 21. 🏦 Loan application analyzer – transform free-text applications into Basel-III risk-model inputs 22. 📊 Earnings call sentiment quantifier – output KPI deltas and investor sentiment scores 23. 💹 Budget variance explainer – produce drill-down pivot JSON for financial analysis 24. 📈 Portfolio risk dashboard builder – feed VaR models with structured investment analysis 25. 💳 Fraud alert generator – map investigation notes to CVSS-scored security metrics 26. 💰 Treasury cash-flow predictor – ingest email forecasts into structured planning models 27. 📊 Financial forecaster – summarize reports into structured cash-flow and projection objects 28. 🧾 Invoice processor – convert receipts into double-entry ledger posts with GAAP tags 29. 📋 Stress test scenario packager – structure regulatory submission data for banking compliance 30. 🏦 Insurance claim assessor – return structured claim-decision objects with risk scores

Healthcare & Life Sciences 🏥 31. 🏥 Patient intake processor – build HL7/FHIR-compliant patient records from free-form intake forms 32. 🧠 Mental health triage assistant – structure referral notes with priority classifications and care pathways 33. 📊 Radiology report coder – output SNOMED-coded JSON from diagnostic imaging narratives 34. 💊 Clinical trial note packager – create FDA eCTD modules from research documentation 35. 📋 Prescription parser – turn text prescriptions into structured e-Rx objects with dosage validation 36. ⚡ Vital sign anomaly summarizer – generate alert reports with clinical priority rankings 37. 🧪 Lab result organizer – output LOINC-coded tables from diagnostic test narratives 38. 🏥 Medical device log summarizer – generate UDI incident files for regulatory reporting 39. 📈 Patient feedback sentiment analyzer – feed quality-of-care KPIs from satisfaction surveys 40. 👩‍⚕️ Clinical observation compiler – convert research notes into structured data for trials

Legal & Compliance ⚖️ 41. 🏛️ Legal contract parser – extract clauses and compute risk scores from contract documents 42. 📝 Court opinion digest – summarize judicial opinions into structured precedent and citation graphs 43. 🏛️ Legal discovery summarizer – extract key issues and risks from large document sets 44. 💼 Contract review summarizer – extract risk factors and key dates from legal contracts 45. 🏛️ Policy impact assessor – convert policy proposals into stakeholder impact matrices 46. 📜 Patent novelty comparator – produce claim-overlap matrices from prior art analysis 47. 🏛️ Legal bill auditor – transform billing details into itemized expense and compliance reports 48. 📋 Case strategy brainstormer – summarize likely arguments from litigation documentation 49. 💼 Legal email analyzer – extract key issues and deadlines from email threads for review 50. ⚖️ Expert witness report normalizer – create citation-linked outlines from testimony records

Customer Experience & Sales 🛒 51. 🎧 Tier-1 support chatbot – convert customer queries into tickets with reproducible troubleshooting steps
52. ⭐ Review sentiment miner – produce product-feature tallies from customer feedback analysis 53. 📉 Churn risk email summarizer – export CRM risk scores from customer communication patterns 54. 🗺️ Omnichannel conversation unifier – generate customer journey maps from multi-platform interactions 55. ❓ Dynamic FAQ builder – structure knowledge base content from community forum discussions 56. 📋 Proposal auto-grader – output RFP compliance matrices with scoring rubrics 57. 📈 Upsell opportunity extractor – create lead-scoring JSON from customer interaction analysis 58. 📱 Social media crisis detector – feed escalation playbooks with brand sentiment monitoring 59. 🌐 Multilingual intent router – tag customer chats to appropriate support queues by language/topic 60. 🎯 Marketing copy generator – create brand-compliant content with tone and messaging constraints

HR & People Operations 👥 61. 📄 CV-to-JD matcher – rank candidates with explainable competency scores and fit analysis 62. 🎤 Interview transcript summarizer – export structured competency rubrics with evaluation criteria 63. ✅ Onboarding policy compliance checker – produce new-hire checklist completion tracking 64. 📊 Performance review sentiment analyzer – create growth-plan JSON with development recommendations 65. 💰 Payroll inquiry classifier – map employee emails to structured case codes for HR processing 66. 🏥 Benefits Q&A automation – generate eligibility responses from policy documentation 67. 🚪 Exit interview insight extractor – feed retention dashboards with structured departure analytics 68. 📚 Training content gap mapper – align job roles to skill taxonomies for learning programs 69. 🛡️ Workplace incident processor – convert safety reports into OSHA 301 compliance records 70. 📊 Diversity metric synthesizer – summarize inclusion survey data into actionable insights

Supply Chain & Manufacturing 🏭 71. 📊 Demand forecast summarizer – output SKU-level predictions from market analysis and sales data 72. 📋 Purchase order processor – convert supplier communications into structured ERP line-items 73. 🌱 Supplier risk scanner – generate ESG compliance scores from vendor assessment reports 74. 🔧 Predictive maintenance log analyst – produce work orders from equipment telemetry narratives 75. 🚛 Logistics delay explainer – return route-change suggestions from transportation disruption reports 76. ♻️ Circular economy return classifier – create refurbishment tags from product return descriptions 77. 🌍 Carbon footprint calculator – map transport legs to CO₂e emissions for sustainability reporting 78. 📦 Safety stock alert generator – output inventory triggers with lead-time assumptions 79. 📜 Regulatory import/export harmonizer – produce HS-code sheets from trade documentation 80. 🏭 Production yield analyzer – generate efficiency reports from manufacturing floor logs

Security & Risk Management 🔒 81. 🛡️ MITRE ATT&CK mapper – translate IDS alerts into tactic-technique JSON for threat intelligence 82. 🎣 Phishing email extractor – produce IOC STIX bundles from security incident reports 83. 🔐 Zero-trust policy generator – convert narrative access requests into structured policy rules 84. 🚨 SOC alert deduplicator – cluster security tickets by kill-chain stage for efficient triage 85. 🏴‍☠️ Red team debrief summarizer – export OWASP Top-10 gaps from penetration test reports 86. 📋 Data breach notifier – craft GDPR-compliant disclosure packets with timeline and impact data 87. 🧠 Threat intel feed normalizer – convert mixed security PDFs into MISP threat objects 88. 🔍 Secret leak scanner – output GitHub code-owner mentions from repository security scans 89. 📊 Vendor risk questionnaire scorer – generate SIG Lite security assessment answers 90. 🏗️ Security audit tracker – link ISO-27001 controls to evidence artifacts for compliance

Knowledge & Productivity 📚 91. 🎙️ Meeting transcript processor – extract action items with owners and deadlines into project tracking JSON 92. 📚 Research paper summarizer – export citation graphs and key findings for literature review databases 93. 📋 SOP generator – convert process narratives into step-by-step validation checklists 94. 🔄 Code diff summarizer – generate reviewer hints and impact analysis from version control changes
95. 📊 API changelog analyzer – produce backward-compatibility scorecards for development teams 96. 🧠 Mind map creator – structure brainstorming sessions into hierarchical knowledge trees 97. 📖 Knowledge base gap detector – suggest article stubs from frequently asked questions analysis 98. 🎯 Personal OKR journal parser – output progress dashboards with milestone tracking 99. 💼 White paper composer – transform technical discussions into structured thought leadership content 100. 🧩 Universal transformer – convert any unstructured domain knowledge into your custom schema-validated JSON

Input Formats

Traditional JSON Input

{
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user", 
      "content": "Your task description here"
    }
  ],
  "context": {
    "session_id": "optional-session-id",
    "metadata": {
      "any": "additional context"
    }
  }
}

YAML Input

messages:
  - role: system
    content: "You are a helpful assistant."
  - role: user
    content: "Your task description here"
context:
  session_id: "optional-session-id"
  metadata:
    any: "additional context"

Template-Based Input

Handlebars Template (template.hbs):

<message role="system">
You are an expert {{expertise.domain}} engineer.
Focus on {{expertise.focus_areas}}.
</message>

<message role="user">
Analyze this {{task.type}}:

{{#each task.items}}
- {{this}}
{{/each}}

Requirements: {{task.requirements}}
</message>

Jinja2 Template (template.j2):

<message role="system">
You are an expert {{expertise.domain}} engineer.
Focus on {{expertise.focus_areas}}.
</message>

<message role="user">
Analyze this {{task.type}}:

{% for item in task.items %}
- {{item}}
{% endfor %}

Requirements: {{task.requirements}}
</message>

Template Variables (vars.yaml):

expertise:
  domain: "DevOps"
  focus_areas: "security, performance, maintainability"
task:
  type: "pull request"
  items:
    - "Changed authentication logic"
    - "Updated database queries"
    - "Added input validation"
  requirements: "Focus on security vulnerabilities"

Structured Outputs with 100% Schema Enforcement

When you provide a --schema-file, the runner guarantees perfect schema compliance:

llm-ci-runner \
  --input-file examples/01-basic/sentiment-analysis/input.json \
  --schema-file examples/01-basic/sentiment-analysis/schema.json

Note: Output defaults to result.json. Use --output-file custom-name.json for custom output files.

Supported Schema Features: ✅ String constraints (enum, minLength, maxLength, pattern)
✅ Numeric constraints (minimum, maximum, multipleOf)
✅ Array constraints (minItems, maxItems, items type)
✅ Required fields enforced at generation time
✅ Type validation (string, number, integer, boolean, array)

CI/CD Integration

GitHub Actions Example

- name: Setup Python
  uses: actions/setup-python@v5
  with:
    python-version: '3.12'

- name: Install LLM CI Runner
  run: pip install llm-ci-runner

- name: Generate PR Review with Templates
  run: |
    llm-ci-runner \
      --template-file .github/templates/pr-review.j2 \
      --template-vars pr-context.yaml \
      --schema-file .github/schemas/pr-review.yaml \
      --output-file pr-analysis.yaml
  env:
    AZURE_OPENAI_ENDPOINT: ${{ secrets.AZURE_OPENAI_ENDPOINT }}
    AZURE_OPENAI_MODEL: ${{ secrets.AZURE_OPENAI_MODEL }}

For complete CI/CD examples, see examples/uv-usage-example.md.

Authentication

Azure OpenAI: Uses Azure's DefaultAzureCredential supporting:

• Environment variables (local development)
• Managed Identity (recommended for Azure CI/CD)
• Azure CLI (local development)
• Service Principal (non-Azure CI/CD)

OpenAI: Uses API key authentication with optional organization ID.

Testing

We maintain comprehensive test coverage with 100% success rate:

# For package users - install test dependencies
pip install llm-ci-runner[dev]

# For development - install from source with test dependencies
uv sync --group dev

# Run specific test categories
pytest tests/unit/ -v          # 70 unit tests
pytest tests/integration/ -v   # End-to-end examples
pytest acceptance/ -v          # LLM-as-judge evaluation

# Or with uv for development
uv run pytest tests/unit/ -v
uv run pytest tests/integration/ -v
uv run pytest acceptance/ -v

Architecture

Built on Microsoft Semantic Kernel for:

• Enterprise-ready Azure OpenAI and OpenAI integration
• Future-proof model compatibility
• 100% Schema Enforcement: KernelBaseModel integration with token-level constraints
• Dynamic Model Creation: Runtime JSON schema → Pydantic model conversion
• Azure RBAC: Azure RBAC via DefaultAzureCredential
• Automatic Fallback: Azure-first priority with OpenAI fallback

The AI-First Development Journey

This toolkit is your first step toward AI-First DevOps. As you integrate AI into your development workflows, you'll experience:

1. 🚀 Exponential Productivity: AI handles routine tasks while you focus on architecture
2. 🎯 Guaranteed Quality: Schema enforcement eliminates validation errors
3. 🤖 Autonomous Operations: AI agents make decisions in your pipelines
4. 📈 Continuous Improvement: Every interaction improves your AI system

The future belongs to teams that master AI-first principles. This toolkit gives you the foundation to start that journey today.

License

MIT License - See LICENSE file for details. Copyright (c) 2025, Benjamin Linnik.

Support

🐛 Found a bug? 💡 Have a question? 📚 Need help?

GitHub is your primary destination for all support:

• 📋 Issues & Bug Reports: Create an issue
• 📖 Documentation: Browse examples
• 🔧 Source Code: View source

Before opening an issue, please:

1. ✅ Check the examples directory for solutions
2. ✅ Review the error logs (beautiful output with Rich!)
3. ✅ Validate your Azure authentication and permissions
4. ✅ Ensure your input JSON follows the required format
5. ✅ Search existing issues for similar problems

Quick Links:

• 🚀 Getting Started Guide
• 📚 Complete Examples
• 🔧 CI/CD Integration
• 🎯 Use Cases

---

Ready to embrace the AI-First future? Start with this toolkit and build your path to exponential productivity. Learn more about the AI-First DevOps revolution in Building AI-First DevOps.