Show HN: Building a Stateful AI Agent (github.com)

A new coding agent named Centri has been announced, designed to enhance AI's memory capabilities by implementing a memory-first approach. Unlike traditional AI agents that forget past interactions and start each session from scratch, Centri features an append-only event spine that acts as a durable source of truth, preserving and accurately recalling every action taken, decision made, and interaction had. This innovative design enables the agent to provide a warm start for every new interaction, eliminating the need for users to re-explain past decisions and minimizing the chances of stale or irrelevant information resurfacing. This development holds significant implications for the AI/ML community, particularly in areas requiring consistent context and historical continuity. Centri’s architecture includes a Python memory API, a typed memory graph with bi-temporal supersession, and deterministic curation with transaction-style receipts, ensuring transparency and traceability of decisions. Additionally, the integration of features like temporal recall, FTS5 verbatim recall, and offline memory consolidation allows for more effective user-agent collaboration and knowledge management, ultimately enhancing productivity in software development and other coding tasks. As Centri builds on the foundations of the open-source OpenCode project, it paves the way for more sophisticated and reliable AI agents that can better serve developers and teams.