Researchers propose PrimeAgentOrchestrator to prime coding agents with personal memory at spawn time

Large language model coding agents typically begin each session with an empty context window, discarding accumulated knowledge from prior work. PrimeAgentOrchestrator (PAO), introduced in a new arXiv cs.AI paper, addresses this by spawning new instances of Anthropic’s terminal-based coding agent, Claude Code, pre-loaded with relevant memories compiled from the user’s existing personal databases.

At spawn time, PAO queries two independently-operated memory backends in parallel: a PostgreSQL entity-observation database and a Cloudflare Worker semantic search index. The system fuses results using backend-specific retrieval strategies and delivers the compiled briefing via filesystem injection that exploits the host agent’s configuration auto-read behavior.

PAO manages the full agent lifecycle, including trust pre-seeding, readiness polling with error detection, and adaptive terminal text injection. The authors report on four months of regular deployment from December 2025 through March 2026 as an experience report, documenting three generations of context delivery mechanisms, the failure modes that motivated each redesign, and the engineering tradeoffs of bridging heterogeneous memory systems rather than building a unified one.