Home

Select a tenant to view the dashboard.

Dashboard

Human Input Needed

Show last SQL attempt 

    

    

    

      
      
      
    

  







  

    
Agents

    
  


  

    
    


      
      

        
Build Organization

        
Start from a proven default template or generate a custom workflow from your data context. The template provides a Data Analyst → SQL Engineer → Operations Agent pipeline that you can later customize with feedback.

        

          
          or
        

        
        

          
        

        
        
      


      
      

        
Give Feedback

        
Describe what should change and the system will auto-detect which agents to update. The workflow structure stays the same — only agent prompts and config are updated.

        
        
        

          
        

        
        

        
      


      
      

        
Re-organize

        
Analyze your agent workflow for overlapping responsibilities, conflicting prompts, and optimization opportunities. Review and apply suggested improvements.

        

          
        

        
        

        
      


      
      

        

          
          
          
          
          

          
          
          
          
          

          
        

        

          

          
          
        

      


      
      

      
      

        
Export / Import Workflow

        
Export your workflow configuration as JSON, or import from a previously exported file.

        

          

            
Export

            
Download the current workflow as a JSON file.

            
          

          

            
Import

            
Upload a previously exported workflow JSON file. This will replace the current workflow.

            
            
            
            
          

        

      


    

  







  

    
MCP Browser

    
  


  

    
    

      

        
        
      

      

        
Select a tenant to load tools.

      

    


    
    

      

        

          
        

        
Select a tool from the left panel to test it.

      


      
    

  







  

    
Review work

    
  

  

    
    

      

        
        
      

      
      

        
Loading…

      

    

    
    

      

        
        
Select a run to view its decision trace

      

      
    

    
    

      

        
        Ask about this run
      

      

        
Select a run, then ask questions grounded in the execution log.

      

      

        
        
      

    

  

  
  

    

      

        
Detail

        
      

      

    

  










  

    
Browse Data Objects

  

  

    
    
  

  

    

      
Warehouse
Select a tenant to browse.

      
      
      
    

    

      
Select a tenant to start browsing the catalog.

    

  








  


    
    

      
      
Ask your data anything.

      
A multi-agent AI platform that turns natural language into accurate SQL and clear business answers — with full traceability and built-in evaluation.

    


    
    

      
Business teams have questions.
Getting answers is slow.

      

        
"What were our top-selling products last quarter?"

        
"Why did customer returns spike in March?"

        
"What's our fulfillment rate by region?"

      

      
Today this means filing a ticket, waiting days for an analyst, and hoping the SQL is correct.

    


    
    

      
Three projects. One platform.

      

        
🧩
Context Builder
Connect warehouses, import catalogs, and define business rules and context

        
→

        
⚡
AI Model
Multi-agent inference engine with workflow design, execution, and review

        
→

        
📊
Eval
Measure quality at every level — clarity, discovery, SQL, execution, accuracy

      

    


    
    

      
Context Builder

      
The single source of truth for your business context

      

        
🏭
Data Catalog
Import warehouses, databases, schemas, tables, and columns from Snowflake or BigQuery.

        
📖
Business Rules
Define tenant rules, domains, glossary terms, guardrails, and calculated columns.

        
🔍
Semantic Search
Vector-powered discovery of tables, columns, and context for accurate SQL generation.

      

    


    
    

      
AI Model

      
A multi-agent pipeline that thinks step by step

      

        
🔍 Analyst

        
→

        
📂 Discovery

        
→

        
💻 Engineer

        
→

        
✅ Tester

        
→

        
📨 Pauley

      

      
Each agent has a specific role. The workflow is config-driven — design it visually, or let AI generate it for you.

    


    
    

      
From question to answer

      

        
🔍
1. Analyst
Interprets the user's question. Identifies intent, required metrics, filters, time ranges, and search queries for table discovery.

        
📂
2. Discovery
Finds relevant tables and columns using semantic vector search and catalog APIs. Fetches business rules for the matched tables.

        
💻
3. Engineer
Generates SQL using the discovered schema, business rules, and guardrails. Handles joins, aggregations, and edge cases.

        
✅
4. Tester
Reviews the SQL for correctness, validates against the schema, and provides feedback. Failed tests loop back to the Engineer.

      

    


    
    

      
Full traceability.
Every decision, reviewable.

      
See exactly what context each agent received, which tools it called, and what decision it made.

      
    
        
  • ✓ Decision Trace — agent-by-agent breakdown with context inputs, tool calls, and outputs
    • 
        
  • ✓ Step-through Review — pause after each agent, edit context, retry, then continue
    • 
        
  • ✓ Human-in-the-loop — automatic escalation with clarifying questions when the AI is unsure
    • 
        
  • ✓ Multi-LLM support — Anthropic, OpenAI with automatic failover and consensus
    • 
        
  • ✓ Full log persistence — every run stored to S3 for post-hoc analysis
    • 
      

    


    
    

      
Puller Eval

      
Measure quality at every level

      
Five-level scoring from question clarity to result accuracy.

      

        
L1 Clarity
95%

        
L2 Discovery
90%

        
L3 SQL Quality
85%

        
L4 Execution
88%

        
L5 Accuracy
82%

      

    


    
    

      
Continuous improvement, not guesswork

      

        
🔍
Test Suites
YAML-based test cases with expected tables, SQL patterns, and result checks. Run full pipeline or context-only.

        
🔄
Model Sweep
Compare multiple LLMs on the same suite. See which model performs best for your data.

        
💡
AI Suggestions
Failed cases get automatic improvement suggestions — new rules, better descriptions, synonym mappings.

      

    


    
    

      
      
Ask your data anything.

      
Multi-agent AI that turns questions into answers.

      
pullerai.com

    


    
    

      
      1 / 10