Conjure Tutorials

View Source

Step-by-step guides for building AI agents with Conjure.

Learning Path

Start with Hello World, then follow the tutorials in order or jump to your use case.

TutorialTimeWhat You'll Learn
Hello World10 minInstall Conjure, create an Echo skill, run your first conversation
Local Skills with Claude30 minBuild a log analyzer skill, deep dive into skill structure
Anthropic Skills API20 minUse hosted execution for document generation (xlsx, pdf)
Native Elixir Skills25 minBuild type-safe skills as Elixir modules
Unified Backend Patterns30 minCombine backends for a complete monitoring solution
Fly.io with Tigris Storage35 minTwo-phase skill pipeline: Claude generates runbooks, Native executes safely

Prerequisites

All tutorials require:

Some tutorials require:

  • Docker 20.10+ (for sandboxed execution)
  • Python 3.8+ (for Python-based skills)
  • Fly.io CLI (for deployment tutorial)

Example Skills

The tutorials use these example skills:

SkillPurposeBackend
echoSimple echo for learningLocal, Docker
log-analyzerProduction log diagnosticsLocal
Native EchoPure Elixir echoNative
Log FetcherREST API log fetchingNative

Use Case: Production Monitoring

The tutorials build towards a complete production monitoring solution:


                    Monitoring Agent                    

                                                        
        
     Native           Local         Anthropic     
     Backend         Backend         Backend      
        
   Log Fetcher     Log Analyzer    Report Gen     
   (REST API)      (Python)        (xlsx, pdf)    
        
                                                        

By the end, you'll have an agent that:

  1. Fetches logs from a REST API (Native backend - fast, in-process)
  2. Analyzes logs with Python scripts (Local backend - shell execution)
  3. Generates incident reports (Anthropic backend - xlsx, pdf)

Use Case: Incident Response Pipeline

The Fly.io tutorial demonstrates a two-phase skill pipeline pattern:


                            Fly.io Machine                               
                                                                         
            
     User       Claude Skill      Runbook Artifact         
    Request         (Anthropic API)       (JSON in Tigris)         
            
                                                                        
                                                                        
                                              
                                              Native Executor Skill    
                                              - Schema validation      
                                              - Action allow-list      
                                              - Safe dry-run           
                                              

Key pattern: LLM generates structured intent → Native skill executes safely

  • Claude reasons and plans: Analyzes incidents, generates structured runbooks
  • Native validates and executes: Schema enforcement, action allow-listing, deterministic execution
  • Artifact-driven: JSON runbook is inspectable, testable, replayable, auditable

Getting Help