Feyn (Chonkie, Inc.)
Feyn builds custom AI models trained on a customer's data, with a production feedback loop that continuously improves the model. Its broader developer-tools work makes AI data ingestion, context engineering, and model ownership faster, cheaper, and more controllable.
At a Glance
- AI application developers and startups
- Enterprise teams building AI-native products
- RAG and knowledge-management teams
- Machine-learning/data-platform engineers
- +2 more
AI Tools by Feyn (Chonkie, Inc.)
(1)
Critic
AI Agent Code Review Tool
Discussions
No discussions yet
Be the first to start a discussion about Feyn (Chonkie, Inc.)
Latest News
Feyn releases MultiMatte, a promptable image matting model built on SAM 3.
Feyn releases FeyNoBg and the NoBg library for background removal and custom training.
Feyn publishes SQRL, which inspects database context before generating queries.
Feyn publishes Pulpie models for fast, cost-efficient web-content cleaning.
Products & Services
An end-to-end service: discover quality metrics, refine an existing AI stack, specialize a model on customer data, and compound improvements through production feedback. Customers own the resulting model weights.
Open-source AI data-ingestion and context-engineering library for cleaning, chunking, embedding/refining, and storing documents for RAG and other AI applications. Available in Python, TypeScript/JavaScript, and Rust ecosystems.
A developer tool for understanding and reviewing code written by coding agents. The website shows GitHub-based code review, local previews, agent questions, and connected-author-device workflows; support is provided by Feyn.
A SIMD-optimized, zero-copy text chunking engine designed for throughput up to 1 TB/s.
Market Position
Feyn positions itself against general-purpose model vendors and heavier or more barebones AI-ingestion tools by combining open-source, lightweight, high-performance infrastructure with managed enterprise delivery and custom models trained on a customer's own data. Chonkie's differentiation is speed, modularity, lower token use (the YC launch claims reductions of more than 75%), and local/on-premise control; the custom-model offering adds continuous production learning and ownership of weights.
Leadership
Founders
Shreyash Nigam
Founder and CEO. Previously worked at Google on on-device AI, ads personalization, and VM memory.
Bhavnick Minhas
Founder and CTO. Builds document-AI and data-ingestion tooling; publicly described as building document AI support at Chonkie.
Executive Team
Shreyash Nigam
Founder & CEO
Previously worked at Google on on-device AI, ads personalization, and VM memory.
Bhavnick Minhas
Founder & CTO
Co-founder of Chonkie and technical lead for document-AI/data-ingestion tooling.
Founding Story
Feyn's founding premise is that a company's data, expertise, and operational knowledge should shape and improve the model it runs rather than merely serve as prompts to a borrowed general-purpose model. The company began in open source with Chonkie, addressing the recurring problem that AI applications fail because their data is disorganized, bloated, or incomplete, and has expanded that expertise into custom models that learn a product's real workflow and improve in production.
Business Model
Revenue Model
Open-source software is free for builders; businesses can buy managed ingestion/custom-model services, white-glove onboarding, hosted deployment, or on-premises deployment. Feyn's custom-model service is positioned as a work-with-us engagement rather than a self-serve plan.
Pricing Tiers
Chonkie and other open-source tools can be installed from PyPI/npm and used in projects.
Managed ingestion pipelines, hosted solution or on-prem deployment, and white-glove onboarding.
Target Markets
- AI application developers and startups
- Enterprise teams building AI-native products
- RAG and knowledge-management teams
- Machine-learning/data-platform engineers
- Companies needing private, self-hosted, or on-premise AI data processing
- Teams building document AI, web-data pipelines, database agents, or image-processing workflows
- Retrieval-augmented generation (RAG) pipelines
- AI-native product data ingestion and context engineering
- Document AI and enterprise knowledge processing
- Custom classification, extraction, and other workflow-specific models
- Web crawling and training-data HTML cleaning
- Database question answering and text-to-SQL
- OpenAI
- Microsoft
- LlamaIndex