Liszt AI Pre-Seed

• We are building Liszt AI.
• Liszt AI makes LLM serving programmable for agentic workflows.
• Today’s inference servers were built for chat: prompt in, tokens out.
• Agents are different. They branch, call tools, retry, verify, and reuse context.
• Forcing agents through a chat API creates extra round trips, repeated prefills, and inefficient KV-cache management.
• Pie lets agent logic run inside the serving system as inferlets.
• Inferlets use workflow knowledge to control KV cache, I/O, and forward passes.
• The result: higher performance, lower latency, lower cost, and better answers.
• Pie is from a research project at Yale CS and first described at SOSP 2025.

What we are raising to build

• Turn Pie into the serving stack for agentic workloads.
• Ship built-in inferlets for common agent workflows.
• Grow the open-source developer and user community.
• Work with teams running agents on open models.

We are looking for investors who can help with

• Introductions to AI infra teams and agent builders.
• Open-source growth and developer adoption.
• Design partners and early customers.
• Hiring strong systems and product talent.

Request a meeting

We are taking pre-seed investor conversations now.

Book an investor meeting