LittlePly: Open-Source Chess Agent Combines 87M Model with Search Harness

Developer andreisavu open-sourced LittlePly, a chess agent built on an 87M-parameter CM-14 model trained from scratch, paired with a GPU-based search harness that can play online. He stresses that model plus harness together form the agent, with clock management and failure recovery also key; phase two will explore learning search via program generation.

2026-11-11 ~ 2026-11-11 · 4 related posts

• LittlePly chess agent wraps act one: 87M model from scratch plus GPU search harness — andreisavu · 2026-11-11
• LittlePly author: exploring and clock management matter as much as the model itself — andreisavu · 2026-11-11
• Chess agent lesson: model plus harness together make the agent, not the model alone — andreisavu · 2026-11-11
• LittlePly open chess agent: 87M from-scratch model, searchable harness, free to play — andreisavu · 2026-11-11