V VibeCI Competitive intelligence
Demo data · public docs
Agents for Business · guided walkthrough

Catch the gap between
a competitor's claims and their docs.

VibeCI reads your strategy, then runs a five-agent pipeline over a competitor's docs — showing where their marketing breaks down, framed for your buyer and grounded to a source line.

Walkthrough

In production VibeCI runs automatically on your behalf — syncing your business context, reading competitors' docs, and producing battle cards with no manual work. The steps that follow simply illustrate what happens behind the scenes.

Your business context · ActivTrak.com Synced & ready
Confluence Messaging pillars Productboard Product roadmap Salesforce Solution map & ICP

These connectors are illustrative for the demo — in production your context syncs live from these tools. It's what steers the agents' research brief.

Pick a competitor to analyze
Analyzing · how it works

Competitor

A look under the hood: the five-agent pipeline producing your analysis, step by step.

0.0s elapsed

Agent pipeline

⚙ under the hood

Five AI agents run in sequence — the first syncs your business context, the rest read the competitor's docs. Each row shows the real tool calls behind the findings.

Source document

⚙ under the hood

The competitor's own technical docs, fetched by the Discovery agent — the text every claim is checked against.

Awaiting Discovery agent… Fetched via the fetch_competitor_docs MCP tool
The ingested documentation will stream in here.
Competitive Intelligence

Competitor

Claims tested against the vendor's own technical documentation.

0
Gaps
0
High severity
100%
Grounded

Claim vs. documented reality

The findings

What VibeCI found for your sales team: each card pairs a marketing claim with what the competitor's docs actually say, plus how to use the gap in a deal.

Source evidence

⚙ under the hood

The competitor's own published documentation. Every finding on the left is grounded to a line in here — click a “View source” citation to jump straight to it. Each gap carries a grounding-confidence score (how strongly the doc backs it), and our eval gates that score in CI.

source.md Competitor's technical documentation · grounding source

Key takeaways

    › Sales battle cardpitch · strengths · objections
    Elevator pitch

    Competitor weaknesses

      Competitor strengths

        Objection handling

        › Discovery questions (sales landmines)
        › Structured output ⚙ under the hood✓ schema validated · CompetitorReport

        The exact JSON the agents produce, validated against the CompetitorReport Pydantic schema before it reaches this screen — the data contract behind everything above.

        
          
        
        
        
      
        
    
        

    
    
        
      
        
        
        
          
        
            Competitive landscape · demo
            
        The whole field at a glance
        
            
        Every rival scored on the capability dimensions your strategy cares about — directed by the same research brief.
        
          
        
          
        
        

        
        
          
        
          
        Anchored in your pillars
        
          
        
        
        

        
        Where ActivTrak wins
        the findings
        
        
        Each cell is the severity of that competitor's documented weakness on that dimension — i.e., your opening. Click a competitor to open their full battle card.
        
        
        
          
        
        
        

        
        
          
        Your openings across the field
        
          
          
        
          
      
          
    
          

    
    
          
      
          
        
          
          
          
            How this works
            
          What just happened
          
            
          VibeCI ran a five-agent pipeline that took your business context, read a competitor's own technical docs, and proved where their marketing overstates reality — every finding grounded to the source line.
          
          
          
          
        
          

        
          
          Kaggle × Google · AI Agents Capstone
          
          VibeCI was built for the “AI Agents: Intensive Vibe Coding” capstone (track: Agents for Business) — a deployable, multi-agent system, vibe-coded by directing a team of coding agents. It doubles as a portfolio piece: a sanitized, public-docs-only replication of a competitive-intelligence workflow I automated at ActivTrak.
          
          
          
            Code on GitHub ↗
            Full write-up ↗
            Watch the 3-min video ↗
          
          
        
          

        
          
          
          
            The workflow
            
          What a team does by hand — compressed into one run
          
            
          Product marketers and sales engineers do this manually: read a competitor's API docs and knowledge base, line them up against the marketing site, find where the claims break, and hand sales a battle card. It's slow, it needs a technical reader, and it's stale the moment the competitor ships an update.
          
            
          VibeCI compresses that into a single pass — ingest the docs, reason about claim-vs-reality, ground every finding to a source line, and emit an account-ready battle card.
          
          
          

          
          
            ⚙ the pipeline
            
          Five specialized agents, each tuned to its job
          
            
            
              
          • Strategy — syncs your business context (pillars · roadmap · ICP) and sets the research brief that directs the rest. The agenda-setter.
            • 
              
          • Discovery — ingests the competitor's docs through an MCP tool server. Low reasoning, high throughput.
            • 
              
          • Technical Analysis ★ — the star: high-reasoning claim-vs-reality contrast. This is the defensible part of the product.
            • 
              
          • Synthesis — structures the findings into a schema-validated battle card.
            • 
              
          • Fact-Checking — high-reasoning QC that grounds every claim back to the source and drops anything unsupported.
            • 
            
          
          
          

          
          
            In an org
            
          Built from real work at ActivTrak
          
            
          This demo illustrates a competitive-intelligence workflow I automated at ActivTrak — rebuilt here on public competitor documentation so it can be shown outside the company.
          
            
          In practice the output feeds sales enablement: battle cards land in the CRM, reps get defensible talking points and discovery questions instead of vague positioning, and PMM stops spending days assembling decks that go stale. The payoff is faster rep ramp, sharper competitive deals, and higher win rates.
          
          
          

          
          
            ⚙ under the hood
            
          The technical build
          
            
            
              
          • Multi-agent orchestration on the google.antigravity SDK (Gemini) — five agents with role-specific reasoning levels.
            • 
              
          • MCP tool server (stdio) exposing the doc-fetch and claims pre-screen tools.
            • 
              
          • Real source grounding — scored & eval-gated — every claim is matched to a line in the ingested doc (no hardcoded citations) and given a confidence score; an eval gates that score in CI.
            • 
              
          • Structured output validated against a Pydantic CompetitorReport schema.
            • 
              
          • FastAPI + Server-Sent Events stream the live run timeline you just watched.
            • 
              
          • Deployed on Google Cloud Run. Demo mode is keyless and deterministic; Live mode runs the real pipeline on a Gemini key.
            • 
            
          
          
          
        
          

        
          
          
          
            
          Built by Daniel Glickman — let's talk
          
            
          I designed and automated this competitive-intelligence workflow at ActivTrak. If you're working on AI agents, product, or sales enablement, I'd love to connect.
          
          
          
          
          
            Connect on LinkedIn ↗
            daniel@cmoconfessions.com