Kev
AI model · Safety / moderation · Text → Tables / structured data by RudderStack
Self-hosted real-time traffic classification and intent detection.
Kev is a self-hosted AI classifier that identifies human, agent, and bot traffic in real-time data streams. It enables businesses to detect malicious intent and categorize automated actors while maintaining full infrastructure control.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Languages
- English
Category: Fraud Prevention
- Model Type
- System 1 Classifier
- Deployment Mode
- Self-hosted private endpoint
- Input Source
- RudderStack Event Stream (Web, Mobile, Server SDKs)
- Output Field
- context.bot object
- Primary Categories
- human, ai_agent, ai_crawler, search_crawler, scraper
- Intent Scoring
- 0.0 to 1.0 probability scale
- Self-hosted System 1 classifier for real-time traffic labeling
- Detects AI agents, crawlers, and scrapers in flight
- Provides 0 to 1 probability scores for bad intent and confidence
- Keeps inference data within your own infrastructure
Overview
Kev is a self-hosted System 1 classifier model developed by RudderStack for real-time traffic identification and moderation. It is designed for businesses that need to distinguish between human users, helpful AI agents, and malicious scrapers while keeping data inference within their own infrastructure. The model runs as a transformation in the RudderStack Event Stream, analyzing request signals to attach bot categories and bad intent probability scores to every event. By providing granular labels rather than a simple binary bot flag, Kev allows teams to serve legitimate AI shopping assistants while blocking harmful scrapers. This infrastructure-native approach ensures that downstream analytics, ad audiences, and security tools receive enriched, trustworthy data without the latency of external API calls.
Detection and Classification
- Traffic Categorization: The model identifies human users, AI agents, AI crawlers, search crawlers, scrapers, and other automation types.
- Bad Intent Probability: Kev provides a probability score from 0 to 1 indicating the likelihood that an event is part of abuse, such as credential stuffing or content theft.
- Kind Confidence Score: The model generates a probability score from 0 to 1 representing the level of certainty for the assigned traffic category.
- User Representation Flag: Kev includes a flag to identify when a real person is behind an automated event, such as a shopping assistant acting for a user.
Infrastructure and Control
- Self-Hosted Deployment: Kev is designed as a self-hosted model, allowing technical teams to point transformations at a private endpoint within their own infrastructure.
- System 1 Architecture: The model functions as a System 1 classifier, optimized for fast, pattern-based decisions on every event in the pipeline.
- Customizable Logic: Data engineers can adjust thresholds, swap model endpoints, or add custom rules within the plain transformation code.
- In-Flight Processing: The model runs as a Transformation in RudderStack Event Stream, classifying events in flight before they land in downstream tools.
Use cases
- Data engineers split product analytics funnels by category to prevent bot traffic from skewing real usage metrics.
- Marketing teams keep scrapers out of retargeting audiences while treating AI agent-driven journeys as a distinct channel.
- Technical executives monitor the volume of AI agent and crawler traffic in their warehouse to plan for automated customer growth.
Specifications
- Access
- Available as a transformation template within RudderStack Event Stream pointing to a self-hosted endpoint
- Model Type
- System 1 Classifier
- Deployment Mode
- Self-hosted private endpoint
- Input Source
- RudderStack Event Stream (Web, Mobile, Server SDKs)
- Output Field
- context.bot object
- Primary Categories
- human, ai_agent, ai_crawler, search_crawler, scraper
- Intent Scoring
- 0.0 to 1.0 probability scale
- Integration Method
- RudderStack Transformation template
How it works
- Collect Events: Events arrive from web, mobile, and server SDKs through the RudderStack Event Stream.
- Classify Traffic: The transformation template sends request signals to the self-hosted Kev model endpoint for a fast verdict.
- Enrich Data: The model writes classification data, including category and badIntent scores, into the event's context object.
- Act on Insights: Users can route, filter, or pass events to over 200 destinations based on the attached labels.
FAQ
Does my data leave my environment when using Kev?
Kev is a self-hosted model, meaning you point the RudderStack transformation at your own private endpoint so request data never leaves your environment.
What types of traffic can Kev detect?
Kev classifies traffic into five main categories: human, ai_agent, ai_crawler, search_crawler, and scraper, plus an 'other_automation' label.
Added by AntSpace from public sources (rudderstack.com).