Overview:
Pixlie AI is an open-source, self-hosted knowledge construction system that builds dynamic knowledge graphs from web content and user-defined objectives. It uses a multi-agent AI architecture to discover, extract, and contextualize information, preserving explicit relationships between entities such as people, places, dates, and events. Designed for developers and researchers who need semantically accurate insights beyond basic vector search, Pixlie runs entirely in the user’s environment with no data leaving the system. It is currently usable for personal web research on a laptop, with a desktop application and enterprise licensing planned for broader use.
Core Features:
Objective-driven knowledge graphs: Users define a goal (e.g., “Track companies on Indian stock exchanges”), and Pixlie autonomously builds a knowledge graph around that objective.
Self-hosted privacy-first architecture: The system runs entirely in the user’s environment; no data leaves the local or cloud deployment.
Multi-model AI agents: Combines large language models and smaller, task-specific models for entity recognition, classification, and relationship extraction.
Web crawling with Brave Search API: Integrates with Brave Search API for web discovery and includes a built-in simple web crawler.
REST API: Exposes a REST API (documented via Bruno in the
rest_apidirectory) for programmatic interaction with the knowledge graph.Entity extraction: Identifies and models people, places, dates, events, and other entities as distinct nodes with explicit relationships (some extraction features are paid).
Use Cases:
Personal web research: Use Pixlie on a laptop to explore and organize information from online sources around a specific topic or question.
Domain-specific monitoring: Set up projects to track entities (e.g., companies, events, individuals) across web content for ongoing analysis.
Prototyping semantic search: Developers can use the REST API and knowledge graph outputs to build apps that require structured, relationship-aware data rather than vector embeddings.
Why It Matters:
Pixlie differentiates itself from vector-only approaches by preserving explicit entity-relationship structures in a self-hosted knowledge graph. It uses a hybrid AI architecture with task-specific agents, allowing users to run the system entirely on their own infrastructure with no data leaving the environment. The open-source core and transparent development make it a practical choice for developers or researchers who need verifiable, structured semantic data from web content without relying on external services or third-party data handling.
