Tsavo Knott

Tsavo Knott is the co-founder and chief executive of Pieces, a developer-tools company building artificial long-term memory across editors, browsers, terminals, documents, and meetings. Its software captures the context developers lose between applications and makes that knowledge available to AI assistants while keeping sensitive work under the user’s control.

Knott graduated from Miami University in 2018 after studying interactive media and computer science. He worked on ventures including Runtime and MeshMyCampus before founding Pieces in 2020 with fellow Miami alumnus Mack Myers. Their initial product helped developers save and retrieve code snippets alongside links, screenshots, images, and text; it subsequently expanded into operating-system-wide context that connects remembered work history, current activity, and AI assistance.

Beginning in late 2022, Knott’s team explored small language models running locally across major platforms, including deployments involving llama.cpp and Ollama. Version compatibility, inference performance, memory consumption, and end-user experience became central engineering problems. A subsequent partnership with Microsoft brought Foundry Local into Pieces’ on-device stack, improving memory management, time to first token, and generation throughput. Foundry Local itself is Microsoft’s platform; Knott’s contribution concerns its application inside Pieces.

• Offline-first AI: Knott treats local inference as a practical foundation for responsive software, lower exposure of sensitive material, and developer tools that remain useful without reliable connectivity.

• Task-specific nano-model architecture: His account of on-device developer productivity describes smaller models handling temporal classification, intent detection, retrieval, planning, and summarization through a coordinated local pipeline.

• Privacy through system design: His writing on AI security focuses on where inference runs, where contextual memory resides, whether prompts or embeddings leave the device, and when users explicitly permit cloud processing.

• Portable context for interoperable agents: Pieces’ Model Context Protocol integration with Cursor makes remembered activity available to external assistants. Its open-source SDK initiative invites developers to build around that context layer, while a native-audio preview extends it into spoken interactions.

Beyond Pieces, Knott has participated in Ohio’s State Committee on Computer Science and supported a Miami University entrepreneurship scholarship.