Balaji Srinivasan presented a vision for personal, private, programmable AI at the SuperAI conference, emphasizing local AI models and local file formats to enable private computation without relying on central servers, according to medianama.com. He highlighted the viral example of combining Obsidian’s local Markdown files with Claude Code, which allows users to query years of notes securely on their own devices.
Srinivasan explained that this approach enables users to make connections across their aggregated historical data without sending information to the cloud. His roadmap involves local AI models, private keys, and encrypted communication within trusted networks. He identified ten forces from AI, crypto, and social media that make this decentralized model feasible, noting that open-weight AI models are only slightly behind closed models in performance.
This vision challenges the current SaaS model by shifting computation and data storage to the user’s device, enhancing privacy and control. It contrasts with centralized AI services that require significant capital expenditure and data sharing. Srinivasan’s approach aligns with broader trends in decentralized technology and cryptographic identity, potentially reshaping how AI tools integrate with personal data.
The example of Obsidian paired with Claude Code demonstrates the practical application of this concept, enabling local queries on private data. Srinivasan’s talk at SuperAI underscores a growing interest in AI architectures that prioritize user privacy and programmability without cloud dependency.