Theoric
An applied AI lab focused on the intersection of formal methods and agentic AI to build fast, bug-free, and performance-correct software.
At a Glance
- AI researchers
- GPU engineers
- AI application developers
- ML Ops
AI Tools by Theoric
(1)
Compute
CLI GPU Cloud Orchestration Tool
Discussions
No discussions yet
Be the first to start a discussion about Theoric
Latest News
Theoric announces agent-convert for Claude and Codex session conversion.
Theoric Labs releases Tgrad, a rewrite of TinyGrad in Lean 4.
Theoric Labs publishes The Clockworks Manifesto.
Harsh Gupta outlines the case for Lean 4 in agentic coding.
Products & Services
A simple CLI interface to use cloud GPUs (RunPod, Hot Aisle, Vast.ai) across providers with one receipt and a flat 7.5% platform fee.
A bounded Lean 4 + Metal runtime for numerical operations (matmul) on Apple Silicon, demonstrating Lean 4 for systems code.
An open-source tool to convert between Claude and Codex session formats for AI agents.
A 'flight simulator' for GPU software that uses hardware modeling and simulation to ensure correctness and performance before running on hardware.
Market Position
Positions itself at the intersection of formal methods and AI, distinguishing from traditional GPU orchestrators (like Modal or RunPod) by focusing on performance correctness and bug-free guarantees via Lean 4.
Leadership
Founders
Harsh Gupta
Founder/CEO of Felvin (pivoted to AI app store), Backend/Chatbots Engineer at HelpShift, Core contributor to SymPy (Nobel Prize winning software). Studied Math/CS at IIT Kharagpur.
Executive Team
Harsh Gupta
Founder & CEO
Previously Founder at Felvin (raised $350K), Senior Software Engineer at HelpShift. Math/CS at IIT Kharagpur.
Founding Story
Started out of frustration with the unreliability of AI coding agents and the extreme complexity of GPU programming. The founder envisioned a future where software is proven correct and performance is simulated at compile-time using formal methods like Lean 4.
Business Model
Revenue Model
Platform fee (7.5% on GPU usage) for Compute.cx; likely enterprise solutions for Clockworks.
Pricing Tiers
Every run uses the provider rate locked at request, plus a flat 7.5% platform fee. Prepaid credit with $10 minimum top-up.
Target Markets
- AI researchers
- GPU engineers
- AI application developers
- ML Ops
- Fine-tuning models
- Reinforcement learning
- Batch inference
- Agentic coding
- GPU software optimization