DEV Community

Manoir Yantai
Manoir Yantai

Posted on

promo_hermes_current

Technical English Variants — v3.5

Dev.to / Medium (long-form)

Just shipped v3.5 of the Memory Sidecar — a production-grade persistent memory system for AI agents. Here's what changed and why it matters.

The problem: Every AI agent forgets between sessions. You spend 45 minutes establishing project context, code conventions, and user profiles. Next session: blank slate.

The sidecar approach: A separate process that runs alongside your agent (Hermes, Claude Code, Cursor, Codex — any of them) and provides 3-tier recall without touching agent internals.

What v3.5 brings:

  • Multi-agent shared runtime — Not just for one agent anymore. Run alongside multiple agents simultaneously, all sharing the same memory pool
  • Bilingual installer — Installer speaks Chinese AND English now. Picks the right language automatically
  • Install modes 1/2/3 — Three structured install paths with automatic fallback. Choose your depth: basic archive → full retrieval → production with all scripts
  • Clear separation — Generic runtime scripts separated from host-specific (Hermes) scripts. Works cleanly with any agent framework
  • KMM ecosystem integration — Official integration with Knowledge-and-Memory-Management. Web scraping → video transcription → document parsing → all flows into the same memory architecture

Memory architecture in one diagram:

Hot Layer (memory tool, 0ms) → Warm Layer (Hindsight PG16, ~50ms) → Cold Layer (gbrain graph + FTS5, ~500ms)
Enter fullscreen mode Exit fullscreen mode

Plus Focused Dossiers: high-priority memory profiles for key people, projects, or recurring incidents. Trigger-based, priority-weighted, timeline-tracked.

One command to install:

git clone https://github.com/mage0535/hermes-memory-installer.git
cd hermes-memory-installer
export AGENT_HOME="$HOME/.hermes"  # or ~/.claude, ~/.cursor
bash install.sh
Enter fullscreen mode Exit fullscreen mode

MIT licensed. 152⭐ on GitHub. 9 scripts for daily operation, 6 more on-demand.

https://github.com/mage0535/hermes-memory-installer

Also co-released: Knowledge-and-Memory-Management v0.0.2 — the knowledge collection pipeline that feeds the sidecar. Web pages, YouTube videos, PDFs, WeChat articles — 40+ tools that autonomously collect, refine, and link into the memory graph.

https://github.com/mage0535/Knowledge-and-Memory-Management

On the vibe coding side: Vibe Coding Universal — 7 rounds of design clarification before writing a single line of code. Matches your answers against 71 real design systems (Linear, Stripe, Airbnb, Apple, etc.) for exact CSS tokens.

https://github.com/mage0535/vibe-coding-universal

Top comments (0)