Syzygy
Syzygy is in a16z speedrun SR005. Based in San Francisco, California, United States of America. Team of 4, founded in 2025.
Facts
- San Francisco, California, United States of America
- 4
- 2025
About Syzygy
Summary
Syzygy, operating under the brand Maniac (maniac.ai), is a San Francisco-based AI infrastructure startup founded in 2025. The company builds technology that allows enterprises to run and own their own task-specific AI models rather than relying on third-party model providers. Its flagship platform, Pioneer by Maniac, orchestrates a proprietary model layer by connecting a customer's live inference pipelines, evaluation tools, and post-training workflows — positioning AI outputs as the customer's own intellectual property. In parallel, the company develops ultra-efficient on-device models under the Mach product line, targeting Apple Silicon hardware, including a 35B mixture-of-experts model that runs on Macs with 16 GB of memory and retains 96.3% of benchmark performance relative to a full-precision teacher model. The broader product suite includes Mach Studio (a desktop app for local model interaction), Mach Engine (a local inference engine for Apple Silicon), and Mach Agents (a TypeScript SDK for building agents with small and local models).
Business model
Syzygy appears to pursue a dual revenue model: a platform or subscription offering for enterprise customers using Pioneer to orchestrate proprietary model layers, and likely a freemium or tiered pricing structure for its Mach Studio desktop product aimed at individual users and developers. The pricing page on the website suggests direct consumer or prosumer monetization alongside potential enterprise contracts.
Go-to-market
Syzygy appears to target two overlapping customer segments: enterprises seeking to internalize and control their AI model stack, and developers or technical users who want capable, efficient models running locally on consumer Mac hardware. The enterprise motion likely involves a direct sales or product-led approach centered on Pioneer, appealing to organizations with data provenance, cost, and customization concerns. The developer and prosumer side is addressed through downloadable Mac software and an in-browser demo, suggesting a bottom-up adoption strategy to build awareness and a user base.
Founders (1)
- Dhruv Mangtani
Score a cohort against your own thesis
GPAgent's sourcing agent reads cohorts like this one and scores each company against your fund's mandate in a private workspace. Your scores, notes, and pipeline never appear on these pages.
Join the waitlist