Harpy logo Harpy

Don't Let Your AI
Fly Blind

Your product's AI sees what users say, not what they actually do.

 

Harpy tracks user behavior in your product and gives your AI richer context for each user (use cases, friction points, engagement level) via a single API call.

You're on the list. We'll be in touch.

Illustration showing how Harpy connects user behavior signals into a unified context graph

Example use cases:

Start where a user left off

Without Harpy: generic one-size-fits-all

What are you trying to do?

📊 Build a new dashboard 🔍 Analyze a funnel or cohort ➕ More

Create, analyze or improve anything

With Harpy: tailored to each person

Last time you got as far as building your first filter — want to pick up where you left off?

🚀 Continue building the filter ➕ More

Create, analyze or improve anything (use @ to reference an Experience or Segment)

The Harpy Payload

Install our snippet into your product, and then make requests via API anytime your AI or agent needs

Harpy
{
  // User Attributes — Properties we generate intelligently and data from your systems via integrations
  "user_attributes": {
    "user_id": "usr_8f3k2m",  "name": "Sarah Chen",  "role": "Senior Analyst",  "plan": "growth",
    "engagement_level": "power_user",  "activity_trend": "increasing",
    "last_active": "2026-03-04",  "...": "more via API or integrations"
  },

  // User Profile — A living overview of who this user is, how they use your product, and where they're headed
  "user_profile": "Experienced power user at Acme Corp, active since January 2025. Relies heavily on reporting and dashboards as core workflows. Recently exploring API capabilities — a signal they may be moving toward deeper technical integration.",

  // Recent Usage — What this user has been doing or focused on over the last 30 days
  "recent_usage_summary": "Over the past 30 days: used reporting daily, created 4 new dashboards, explored API settings twice without completing setup, managed team permissions weekly, and has not engaged with integrations at all.",

  // Recent Sessions — What happened in their most recent visits in your product
  "recent_sessions": [
    { "date": "2026-03-04", "mins": 20, "summary": "Opened reporting, duplicated a dashboard, customized filters. Navigated to API settings, viewed auth page for 90 seconds, then left without taking action." },
    { "date": "2026-03-03", "mins": 15, "summary": "Went to Team Admin, reviewed permission roles for two members. Switched between roles settings and help docs three times. Updated one role, left the other unchanged." }
  ],

  // Friction Signals — Where this user experienced issues or failed to succeed
  "friction_signals": [
    { "feature": "API > Auth",   "signal": "Visited 2x, exited",       "severity": "medium" },
    { "feature": "Team > Roles", "signal": "Back-and-forth with docs", "severity": "low"    }
  ],

  // Feature Adoption — Which features they've explored, how deeply, and how often
  "feature_adoption": {
    "reporting":    { "depth": "advanced",  "frequency": "daily"  },
    "dashboards":   { "depth": "advanced",  "frequency": "weekly" },
    "integrations": { "depth": "none",      "frequency": "never"  },
    "api_settings": { "depth": "exploring", "frequency": "recent" }
  },

  // Company Context — How the whole account is using your product across all users
  "company_context": {
    "company": "Acme Corp",  "plan": "growth",  "total_users": 34,
    "summary": "Acme Corp has 34 active users, up from 20 three months ago. The team is heavily invested in reporting and dashboards. No users have adopted integrations yet. Three users besides Sarah have explored API settings, suggesting growing technical appetite across the team."
  },

  // Personalization Directive — Direct instructions for your agent on how to personalize for this user
  "personalization_directive": "Address Sarah as a technical, experienced user — skip introductory explanations entirely. Use precise language. If she asks about APIs, acknowledge she's already started setup and guide her from the authentication step forward. Proactively recommend integrations as her highest-value untapped area. Do not suggest reporting features — she knows them better than most."
}

From the team that built Chameleon, a product vetted by hundreds of engineering and security teams to run in production.

Our behavioral intelligence infrastructure already processes tens of millions of sessions monthly.

0B+

Sessions

99.9%

Uptime SLA

SOC 2

Type II Certified

We're onboarding fast moving teams with AI or Agents in production.

You're on the list. We'll be in touch.

We built behavioral tracking infrastructure for:

ClickUp DRATA Fivetran StackBlitz
How it works 1 How it works 2 How it works 3 How it works 4

How long does setup take?

Install the snippet in minutes, and supplement with integrations for additional attribute or event data. Personalization context will be generated after 24 hours and will improve over time.

 

Can I send data from other sources?

We support modern data stack, including CDPs (e.g. Twilio Segment), reverse ETL (e.g. Hightouch), CRMs (e.g. Hubspot), analytics providers (e.g. Amplitude) and manual (webhooks, APIs)

 

What data do you store?

Harpy tracks user activity (navigation, interactions etc.) and processes into summaries and traits, which are stored. We avoid capturing PII or sensitive data (e.g. input fields).

 

Does this work with any LLM?

While we use OpenAI models, you can use our API with any LLM or Agent that you are building, regardless of provider.

 

How do I actually implement or use Harpy?

You can choose where and when you'd like Harpy to provide your AI or agent its rich user context. At that point you can make an API call to Harpy to return the payload. More in our developer docs.

 

Does this support account-level context?

Yes. Harpy tracks activity across all users in a company, generates a company-level usage summary, and folds that into each user's personalization directive. This means your agent can factor in what the broader team has done — not just the individual — when forming its response.

 

What does pricing look like?

We charge based on API calls to tie into the value you're receiving. Favourable pricing and credits available for our early adopters.

 

Do you have an MCP server?

An MCP Server for Harpy is not available yet, so you cannot yet add it directly to an LLM and ask about your users, but that's on the roadmap.