Changelog

What's new in GoldHold. We ship fast.

v19.0

Latest February 20, 2026

🗄️ GoldHold Vault

Binary file storage — upload, download, search, sync, list, and stats
Cloudflare R2 backend — free tier (10GB storage, zero egress)
Semantic search for files — find anything by describing it
Setup choice: "local only" or "Cloudflare R2" during install
Bulk sync for entire directories

🏷️ Rebranded Hooks

boot-md → GoldHold-boot — human-friendly descriptions
session-memory → GoldHold-journal
session-save → GoldHold-save

🔧 Version Unification

All scripts share v19.0 — no more version drift between components

v18.0

February 18, 2026

📉 Token Burn Fix

SOUL/AGENTS/HEARTBEAT patches reduced from ~4KB to ~800B
Patches are now pointers to docs, not content dumps
Run --patch-only to get slim versions

🧹 Root Cleanup on Install

--upgrade auto-cleans workspace root
Moves reference docs to memory/core/
Fixes VBS watcher paths automatically

v17.0

February 18, 2026

🧠 Intelligence Layer

Context Diet Mode — automatic context window optimization
Model Profiles — per-model configuration
GitHub auto-access for subscribers
Persona drift detection and auto-correction

v16.0

February 18, 2026

👁️ Watcher & Autostart

memory_watcher.py — persistent file-watching event loop
Windows autostart wrapper (launches silently on login)
--doctor now checks watcher health

v15.0

February 17, 2026

⚡ Zero-Friction Install

One-paste setup — paste API key, everything else automatic
Browser auto-opens to cloud signup
Existing indexes auto-detected and reused
Multi-agent security with role-based write permissions
9-check diagnostic with --doctor
Health dashboard with letter grades (A/B/C/D/F)

v14.0

February 17, 2026

🛡️ Universal Hardening

Pre-compaction flush — agent saves state before context is cleared
4 session hooks (save, boot, transcript, command logging)
13-check health monitor with auto-remediation
Crash recovery with conversation replay on boot
30-day receipt auto-archival

v13.0

Initial Release February 14, 2026

🚀 Launch

Cloud-backed semantic memory — zero models, zero GPU, zero Docker
24 CLI commands for sync, search, backup, and recovery
Three-layer crash survival: cloud + git + local export
Multi-agent mode with boss/worker roles
One-command installer with auto-config
Conversation capture with dedup

7 releases in 6 days. We're just getting started.