§1 — Opening
The past few months with Hermes Agent (by Nous Research) have been nothing short of remarkable. It recently surpassed Claude Code in GitHub stars and continues to rank among the most active and fast-growing repos!
What stands out even more is the pace at which the Nous team ships new features, and of course the vibrant community forming around it. And that’s what Hermes Atlas is all about - helping you stay on top of everything that’s going on!
This is the first issue of the Hermes Atlas newsletter. The plan is to publish every couple of weeks, covering product updates, community highlights, and interesting projects and use cases. If you have any suggestions or anything you’d like to suggest to be included just hit me up on X.
§2 — Hermes product updates
Hermes Agent v0.12.0 — "the curator release" (April 30)
The biggest release yet — 1,096 commits and 213 community contributors since v0.11.0. The headline is the autonomous curator: a background agent that runs on a 7-day cycle, grades your skill library by usage, prunes dead skills, and consolidates duplicates without you asking. The self-improvement loop that decides what to save also got a rubric-based
rewrite to class-first grading instead of free-form "should we save this."
A few other notables:
comfyui v5 + touchdesigner-mcp — both moved from optional to bundled-by-default
Microsoft teams — 19th messaging platform, shipped as the first plugin-platform
native Spotify + Google Meet — join a meet, transcribe, follow up
LM Studio first-class provider — fully local setups got a real upgrade
57% cut to TUI cold start — visible on every launch
Karpathy's LLM wiki skill got a polish pass in 4-23 (provenance markers, source hashing, quality signals)
→ full release notes: here
§3 — Atlas updates
11 new projects were added to the ecosystem map since the initial launch, a few notables include:
mem0 — an intelligent memory layer that fuses semantic search, BM25 keyword matching, and entity linking to recall user preferences and session history
Honcho — a configuration layer for self-hosting Plastic Labs' Honcho memory backend (I personally use Honcho)
ContextPilot — inference optimization layer designed to accelerate long-context LLM workloads by maximizing KV cache reuse
portable-hermes-agent — a Windows-native desktop application that provides a user-friendly interface for the Hermes Agent framework without requiring Docker
super-hermes — opinionated skill bundle that enables models to generate their own analytical frameworks, or prisms, for deep code and document inspection
agent-browser-mcp — MCP server that enables AI agents to interact directly with a user's active Chrome browser session rather than a sandboxed environment
hermeshq — a Docker-first control plane designed to orchestrate and manage multiple Hermes Agent instances from a single web application
Two new site features also shipped:
Featured Product section — a rotating community spotlight on the homepage where each week a new project will get the spotlight
→ browse the full atlas: https://hermesatlas.com
§4 — Spotlight: Hermes Workspace
A native web command center for the Hermes ecosystem — and arguably the cleanest GUI anyone has built for it so far.
What it does:
multi-model chat + live tool execution
memory browser (search & edit agent context inline)
skills catalog (100+)
swarm mode — multi-agent control plane with role-based dispatch and a kanban board
built-in terminal — super convenient
mobile-first PWA — full feature parity on desktop, tablet, or phone
The pitch is one interface for the parts of your agent that usually scatter across terminals, tabs, and logs and it connects directly to a vanilla hermes-agent install.
→ for more, checkout: https://hermes-workspace.com/
§5 — Featured use case: a Hermes Agent to run your Karpathy LLM wiki “second brain”
We’ve all seen Karpathy’s LLM wiki post that has now exceeded 20M views, and many people have likely tried, or wanted to try, building this for yourself. The only friction is that you need to maintain it, and that’s a perfect use case for a Hermes Agent.
Jsong (@jsong_49820) built exactly that - an agent that runs the LLM wiki and makes it accessible via web for access anywhere. The stack:
a Hetzner VPS (2 vCPU / 4 GB) running Hermes 24/7
Telegram as the front door
Karpathy's LLM Wiki pattern as the storage substrate
a static-site frontend to make the wiki browsable on the open web (no need for Obsidian)
Why they did it: Obsidian had three limits they couldn't solve — maintenance debt (manual cross-references), desktop lock-in with the Obsidian app, and notes that never improved on their own.
The Hermes-managed wiki reads incoming messages, files them under the right page, cross-links to existing concepts, and audits for stale claims and orphan pages via crons. The intelligence compounds instead of rots.
The kicker: this exact pattern ships as a bundled llm-wiki skill in Hermes now — and got a meaningful upgrade in the 4-23 release (provenance markers, source hashing, quality signals) so all you need to do is use it and decide what frontend you want.
→ full writeup: here
§6 — Closing
That's issue 01. We’ll be back with the next issue in a few weeks.
If something here was useful, two things help us a lot:
→ drop a ★ on the atlas repo (helps with discovery): https://github.com/ksimback/hermes-ecosystem
→ reply with one project we should feature next, or a build of your own we should write up
— Kevin Simback
maintainer · x.com/ksimback