Hermes Agent Masterclass: 3. Memory, Plugins, Honcho, and Obsidian

Module 3 of the Hermes Agent Masterclass — opening up Hermes' built-in memory system and stacking three more layers on top of it. Sign up for my FREE weekly newsletter, where I spill my unfiltered thoughts on the latest AI news, cool research, and projects I'm building: https://www.onchainaigarage.com/ 🐦 Follow Tonbi on X for real-time AI x blockchain updates! https://x.com/tonbistudio Modules 1 and 2 got Hermes running, sandboxed, always-on, and reachable from your phone via Telegram and Discord. What you might not have realized is that Hermes' memory system has been quietly doing work the whole time. This module opens it up: how the built-in markdown memory actually works, the frozen-snapshot pattern that lets it stay cache-friendly, the four invisible safety features (cap-as-feature, save/skip policy, silent dedupe, injection scanning) that keep it from collapsing into a junk drawer, and then three more layers that stack on top — FTS5 session search over every prior conversation, the pluggable memory provider ecosystem (Honcho, Mem0, Hindsight, SuperMemory and more), and the Obsidian skill for long-form structured knowledge. Ends with a working setup of Honcho on a real agent and the Obsidian skill writing a multi-file vault for the HVAC-quotes project we've been building across the masterclass. ✅ Layer 1 + 2 deep dive — memory.md and user.md (the two markdown files in .hermes/memories), how they auto-load into the system prompt at session start, the frozen-snapshot pattern (writes hit disk live but only show up in the next session, so prompt caching stays valid), the three-action memory tool (add/replace/remove, no read), and the four design choices that make a 1,300-token cap actually work. Plus FTS5 session search across every prior CLI/gateway conversation in state.db, with the v0.11+ auto-prune+vacuum on startup so the DB self-maintains. ✅ Layer 3 — provider plugin landscape and a full Honcho walkthrough. Comparison of the main options (Mem0's server-side LLM extraction, Hindsight's knowledge graph + Reflect synthesis, SuperMemory's multi-container partitioning, and Honcho's dialectic user modeling), then `hermes memory setup` end-to-end against Honcho cloud — workspace/peer/session hierarchy, the four Honcho tools (conclude, context, profile, search), recall and write-frequency config, and a live test where the agent stores and recalls a name through the Honcho conclude path. ✅ Layer 4 — Obsidian skill (not a plugin, a bundled skill at skills/note-taking/obsidian) writing real Wikilink-cross-referenced markdown into an actual Obsidian vault. Used it for the HVAC-quotes project from the previous module: ten Bay Area HVAC providers researched into per-company notes plus an equipment-models index, all queryable by the agent on future sessions. Good fit for project-scale long-form knowledge that's too detailed for the 1,300-token markdown cap. 💻 Tonbi's GitHub: https://github.com/tonbistudio 🌐 Portfolio: https://www.tonbistudio.com Resources: 🔗 Honcho (cloud): https://app.honcho.dev/ 🔗 Nous Research: https://nousresearch.com/ 🔗 Obsidian: https://obsidian.md/ 🔗 Mem0: https://mem0.ai/ Timestamps: 0:00 - Intro and the four memory layers 1:30 - Layer 1: memory.md and user.md walkthrough 6:00 - Frozen snapshot pattern and the four safety features 12:00 - Layer 2: FTS5 session search and auto-prune 15:00 - Provider plugin landscape: Mem0, Hindsight, SuperMemory 21:00 - Honcho setup, dialectic reasoning, live test 26:00 - Layer 4: Obsidian skill on the HVAC project 33:00 - Wrap up and what's next Coming Next: Module 4 — Skills. Memory is what Hermes knows; skills are what Hermes does. We'll dig into the bundled skill catalog, the custom-skill workflow that lets your agent grow new capabilities over time, and continue building out the HVAC enhanced-quotes project with skills wired in. 👀 How are you handling memory in your agent setup — built-in markdown, a provider plugin, an Obsidian-style vault, or something else entirely? If you're running Honcho, Mem0, Hindsight, or SuperMemory in production, drop your verdict in the comments. If this masterclass series is useful, please like, subscribe, and hit the bell for Module 4! 🦐✨ #HermesAgent #HermesMasterclass #AIMemory #Honcho #Obsidian #Mem0 #LocalLLM #AITools #VibeCoding