ESAA-Conversational: Event-Sourced Memory for Seamless LLM Agent Handoffs

Software developers frequently juggle multiple LLM coding agents like OpenAI Codex, xAI Grok, and Anthropic Claude Code, but each tool keeps its conversation in a private, vendor-specific log. This causes conversational state drift—goals, decisions, and open tasks established with one agent are lost when another takes over. ESAA-Conversational, a specialization of the Event-Sourcing Agent Architecture (ESAA), solves this by treating the visible conversation as a local event store. It uses hooks and watchers to mechanically capture turns, normalize them into an append-only log, and deterministically project read models such as active goals, decision records, task lists, and context summaries—all without requiring LLM inference for capture. Agents only use judgment for explicit curation via domain commands like `decide` and `task`.

The public v1.1.0 release implements a PowerShell CLI with eight commands (`init`, `enable-hooks`, `sync`, `project`, `verify`, `context`, `decide`, `task`), `workspace_root` isolation, and a write-path lockfile. It's distributed as a greenfield package with an empty public log. A self-referential case study with 570 development-lab events shows that heterogeneous agents can collaborate through a shared log without a direct agent-to-agent channel. Privacy is preserved by excluding private conversational history from the public distribution. This approach enables seamless multi-agent workflows, preventing loss of context and decisions when switching between AI coding assistants.