Show HN: EchoMode – A stability layer that prevents persona drift in LLMs (github.com)

Echo Mode is an open‑core protocol layer that adds a deterministic “spine” to LLM-driven assistants: a finite‑state machine (FSM) plus transparent heuristics to constrain and stabilize persona, tone, and multi‑session behavior. The OSS v1.3 package (TypeScript, Node 18+, pnpm, Apache‑2.0) ships an FSM core (@echo/fsm), baseline scoring heuristics (@echo/heuristics), an Express middleware for API‑level enforcement (@echo/middleware), a React HUD (@echo/hud) and CSV/JSONL exporters for analytics. The protocol exposes named states (Sync, Resonance, Insight, Calm) and deterministic transitions so developers can detect drift (scores) and programmatically transition or recover to target behaviors. Telemetry is off by default, all heuristics run client‑side, and the repo documents the OSS ↔ commercial split: advanced calibration weights, dashboards, connectors and SaaS control live in the closed commercial repo. For the AI/ML community this matters because it tackles a practical pain point—emergent persona drift and inconsistent multi‑turn behavior—using interpretable FSM rules rather than opaque model fine‑tuning. That makes assistant behavior more predictable, auditable, and recoverable across sessions and providers, improving governance, reproducibility, and analytics. Limitations: the OSS bundle contains only heuristic scoring (no ML weights), and more sophisticated calibration and monitoring features require the commercial offering. Overall Echo Mode is a pragmatic layer for teams building long‑lived, multi‑session agents who need stable, enforceable conversational personas.