elsechord

Infrastructure for software
and embodied agents.

We build the infrastructure agents need to understand environments, learn tasks, and act in software and the physical world.

From navigation research
to agent engineering.

Elsechord was initiated by Peichen Chen. At Hangzhou Dianzi University, he studies vision-and-language navigation and world models while building multi-agent workflows and tool systems. This experience informs how we approach environments, tasks, and feedback together.

Meet the founder

Agent capabilities should
grow through work.

We believe progress will increasingly depend on what agents experience and how they learn from it. We connect memory, task environments, and execution systems so experience can inform the next round of training and action.

Research and technical reports

Understanding that
changes with experience.

Later scenes in a film can change the meaning of earlier clues. We use long films to study memory, active revisiting, and revised interpretations, building tools for indexing, retrieval, and review.

Long-form understanding and review

Turn real work
into learning environments.

Starting with Chinese business data, we build benchmarks for querying, verifying, and making business judgments, then develop interactive environments where feedback from real tasks can support training.

Tasks and learning environments

Connect understanding
to action.

Real tasks require tools, coordination, and feedback. CyberGuard is our open-source work in this direction. We are exploring replayable, analyzable execution histories that connect training and runtime.

Agent systems and open-source work

Build agent infrastructure
with us.

We want developers to build agents that operate across software and the physical world. Join us in research and engineering, or bring a real task.