Skip to content

Memory Systems for Agent Fleets

A research mini-site on memory systems for long-running multi-agent fleets: local-first runtime design, sync contracts, explicit identity and async semantics, and federated failure modes.

Papers

  1. Designing a Local-First Memory Runtime for Agent Fleets — field study on moving from a heavy backend stack to a single edge process.
  2. A Taxonomy of Sync Strategies for Fleet Memory — decision framework for fleet-memory sync contracts.
  3. Explicit Identity and Async Semantics in Agent Memory Tools — tool-design argument for memory surfaces.
  4. Failure Modes in Federated Agent Memory — incident experience report on federated-memory contract failures.

Artifacts

Research process

The operational foundation for this research is documented in the Agent Fleet Research site, which studies multi-agent operations fleets and role-based collaboration.

Building this site

cd /home/yakov/memory-research-site
.venv/bin/mkdocs serve
.venv/bin/mkdocs build --strict