Internal Strategic Overview · For Roland

Two agents, one origin,
divergent evolution.

Mark-1 and the stock Hermes Agent share the same codebase. Over the past weeks they have evolved in different directions — Hermes hardens a general-purpose platform for everyone; Mark-1 has been built upward into a single executive's operating system. This is the map of that divergence, measured exactly.

Source of truth: the public NousResearch/hermes-agent repo on this VPS · pinned at v2026.6.5 + 301 commits · HEAD 4219a91df · all figures below counted directly from disk.

170
Hermes skills available
(75 core + 95 optional)
193
Mark-1 active skills
104
Net-new · built for Mark-1
16
Net-new domains
Scroll — the full divergence below
The Core Thesis

Same codebase. Opposite evolutionary pressure.

Both systems run the identical Hermes Agent engine (v2026.6.5, the v0.16.0 line). What separates them is not the engine — it is what each is being optimised for. Hermes optimises for breadth across millions of unknown users: it ships 75 core skills enabled by default plus 95 optional ones you can switch on. Mark-1 optimises for depth for one known operator and one business: it keeps every relevant upstream skill and adds 104 of its own.

Upstream Hermes Agent

Horizontal — a platform

Vector: stability · breadth · safety for the median user
  • Hardens the runtime. The 301 commits since the v0.16.0 tag are memory-leak fixes, gateway stability under load, PTY teardown, plugin discovery, voice-during-run transcription.
  • Ships a wide catalogue — 75 core + 95 optional skills — most disabled by default, covering what an average user might want.
  • Stays generic. No memory of who you are, no business context, no standing goals. Every session starts cold by design.
  • Serves everyone equally — which means it serves no one deeply.
Mark-1

Vertical — an operating system

Vector: depth · context · leverage for Roland & Nionium
  • Builds upward. The 104 net-new skills include whole subsystems — a SaaS product, a knowledge brain, an autonomous intelligence layer, a self-correcting loop discipline.
  • Carries persistent context. Knows the portfolio, the properties, the family, the businesses, the standing positions — and applies them unprompted.
  • Acts autonomously on a schedule against your goals, then verifies its own work before reporting.
  • Serves one operator deeply — which is the entire point.

Read plainly: upstream Hermes is getting more reliable and broader. Mark-1 is getting more yours. Both matter — Mark-1 inherits every upstream stability gain on each safe update, then layers its own vertical stack on top.

The Honest Math

Where Mark-1's 193 skills come from

Not all 193 are bespoke — and claiming so would be dishonest. Here is the exact composition, counted by set intersection against the upstream repo. 89 skills are inherited verbatim from Hermes; 104 were written specifically for Mark-1.

193 active Mark-1 skills, by origin

Each segment is proportional to its real count.
89 inherited
59 net-new subsystems
45 net-new integrations
89 inherited — kept verbatim from upstream (all 75 core + 14 optional)
59 net-new subsystems & infra — deep engineering (DataDesk, Brain, intelligence, ops)
45 net-new integrations — data sources & content workflows

For symmetry: upstream also ships 81 optional skills Mark-1 has not enabled (blockchain, large parts of the mlops catalogue, finance widgets, etc.) — capability that exists in the platform but isn't relevant to this operator. Breadth is not the same as fit.

The Fork

Where the two paths split

From a shared origin, each system's recent work pulls in a different direction. Left: what upstream actually shipped (verifiable in the git log). Right: what was built into Mark-1 over the same window.

Shared origin · Hermes Agent v2026.6.5 · agent loop, tool system, gateway, skills framework, memory primitive, platform adapters
Hermes → platform hardening
301 commits since v0.16.0 · all in the public log
git logGateway memory-leak fixes — release evicted agent clients, close residual leak sites under scheduled load
git logPTY process-group teardown — clean termination of interactive subprocesses
git logNested-category plugin discovery + alias-normalized enable/disable
git logVoice transcription during active runs; heap-dump disk-fill guards
git logOff-thread end-of-turn memory sync; interrupt-hang fixes
Mark-1 → vertical capability
104 net-new skills · whole subsystems on top
+16 skillsDataDesk SaaS — chat product, voice, billing, admin, multi-tenancy on the agent
+3 skillsSecond Brain — durable knowledge graph, hybrid retrieval, nightly dream-cycle
+6 skillsAutonomous Intelligence — scheduled sweeps reasoning over feeds against your goals
+4 skillsIntel Map + Iran Daily — live geospatial OSINT & daily SITREP synthesis
+1 skillLoop Discipline — goal-aware stop + mandatory self-verification on unattended runs

The asymmetry is the message. Upstream's column is fixes to the shared engine. Mark-1's column is new organs grown on top of it. Neither replaces the other — Mark-1 keeps pulling the engine fixes in via safe updates (the hermes-safe-update skill exists precisely for this).

The Additions · Why & Benefit

16 net-new capability domains

These are skill domains absent from the upstream repo entirely — built specifically for Mark-1. For each headline subsystem: what it is, why it was implemented, how it benefits you, and exactly what stock Hermes does in its place.

DataDesk — the product

16 net-new skills · chat · voice · billing · admin
Why built

To turn the agent into a sellable "AI for Executives" SaaS at $980/mo — a real revenue line for Nionium, not an internal tool.

Your benefit

A commercial asset you own end-to-end: chat UI, Stripe billing, invite codes, voice mode, admin dashboard, per-tenant isolation. Mark-1 is both the operator and the product.

Stock Hermes: a CLI + gateway. No product layer, no billing, no multi-tenancy — by design.

Second Brain

3 net-new skills · SQLite · hybrid search · dream-cycle
Why built

Upstream's memory is a flat string injected per turn. An executive needs a connected, queryable knowledge base — pages, entities, relationships — that compounds across sessions.

Your benefit

Ask "what do we know about X" and get a synthesised, cited answer from everything ever captured — with an honest "gaps" section. Cross-references your goals automatically; self-maintains nightly.

Stock Hermes: a single compact memory string per turn. No graph, no retrieval, no synthesis.

Autonomous Executive Intelligence

6 net-new skills · scheduled sweeps · tiered signals
Why built

You shouldn't have to ask. The agent should ingest feeds daily, reason about them against your strategic goals, and surface only what creates or removes an opportunity or risk.

Your benefit

An analyst that works while you sleep — AM sweep + PM watch, tiered alerts (urgent / digest / silent), strictly inside a hard money boundary it can never cross without you.

Stock Hermes: reactive only. Has a cron tool, but no standing goals and no autonomous reasoning layer.

Intel Map & Iran Daily

4 net-new skills · MIL-STD-2525 · live OSINT · SITREP
Why built

Geopolitics (Iran, Ukraine, Georgia) directly affects your portfolio and decisions. You wanted a live theater map and a daily intelligence product, not a news feed.

Your benefit

A self-updating military-standard map (forces, flights, vessels, pipelines, cables) plus an original daily SITREP in Mark-1's own analytical voice — published, archived, shareable.

Stock Hermes: can fetch a URL via the browser tool. No geospatial layer, no recurring synthesis pipeline.

Loop Discipline

1 net-new skill · goal-aware stop · self-verification
Why built

Unattended runs were either grinding to the turn cap or reporting "done" when nothing actually changed. The loop needed to stop for the right reason and prove its own work.

Your benefit

Autonomous jobs self-correct: they halt when stuck (and say why), and refuse to claim success until reality is independently verified. Fewer false "done"s reaching you, zero new approval prompts.

Stock Hermes: a hard max_turns ceiling and task-completion stop. No goal-aware halting, no mandatory verification stage.

Hermes Self-Operations

8 net-new skills · safe-update · gateway · cloning
Why built

Mark-1 must maintain itself: pull upstream fixes without losing the vertical stack, tune its own gateway, clone instances, debug its own TUI, register custom providers.

Your benefit

The meta-layer that keeps Mark-1 maintainable. This is precisely how Mark-1 inherits all 301 upstream commits while keeping its 104 bespoke skills intact — no fork drift, no lost work.

Stock Hermes: ships the update mechanism, but no codified safe-update / clone / gateway-ops playbooks.

Realtime Voice & Avatar

4 net-new skills · OpenAI Realtime · HeyGen
Why built

You want true sub-second voice interaction (and a live avatar) tuned for an executive persona — not the generic voice path the platform ships.

Your benefit

Hold real-time voice conversations in the Marin voice with three-layer noise suppression, plus an optional live video avatar — desk or pocket.

Stock Hermes: has voice adapters, but no tuned Realtime executive persona or avatar pipeline.

Messaging & Telegram Ownership

5 net-new skills · gateway · multi-instance · webhooks
Why built

You operate from your phone via Telegram. Running multiple Mark-1 instances on one bot token required solving polling-conflict ownership and webhook routing.

Your benefit

Mark-1 reaches you reliably on Telegram (with Mari read-only), delivers the morning briefing there, and the hub-vs-spoke ownership model is documented so spokes never hijack the bot.

Stock Hermes: ships a Telegram adapter, but no multi-instance ownership model or conflict resolution.

Beyond these subsystems, the remaining net-new skills are 45 data-source integrations and content workflows — detailed in the next section — the breadth that lets Mark-1 answer from real data instead of generic knowledge.

The Integration Layer

36 integrated external APIs

Stock Hermes ships generic search and a browser. Mark-1 is wired directly into authoritative live data sources, so it answers from the actual numbers — markets, filings, science, geodata — not from model memory. This is the data hierarchy in practice: check the integrated source first, web search only as fallback.

6 categories 36 live integrations FMP → FRED → DBnomics finance order Tavily → Exa web order
Financial & Markets 6
Primary source for any money question — checked before web.
FMPStocks, forex, crypto, fundamentals
FRED816K+ US/global economic series
DBnomics1.7B macro time series, aggregated
World Bank17K+ country development indicators
SEC EDGARUS filings — 10-K/10-Q/8-K, full-text
PolymarketPrediction-market prices & odds
Search & Web Intelligence 6
Real-time discovery, scraping, and OSINT corpus.
TavilyFirst-choice real-time web search + images
ExaNeural/semantic search, findSimilar
FirecrawlScrape / crawl / extract structured data
Felo SearchSearch + X/Twitter neural search
Context7Up-to-date library/framework docs
VERIDIUM (CustomGPT)14.7K-doc geopolitics OSINT corpus
Scientific & Academic 5
Primary literature and structured science data.
PubMedBiomedical & life-sciences literature
OpenAlex250M+ academic works, all fields
PubChemChemical compound data
UniProtProtein sequence & function
arXivPreprints by topic / author / category
Travel, Local & Lifestyle 6
Booking, geodata, and day-to-day utility.
DuffelFlights — 300+ airlines, book & manage
Booking.comHotel search with star/area filters
Open-MeteoWeather — current, hourly, daily
OSM / OSRMGeocode, POIs, routes, timezones
SpotifyPlayback, search, queue, playlists
MatsneGeorgian legislation retrieval
Media & Content 5
Feeds, sports, encyclopaedic media.
YouTube innerTubeChannel/video data (EU-region bypass)
TheSportsDBLive scores, fixtures, squads
Wikipedia / CommonsArticles + license-clean images
TenorGIF search
Felo MindmapMindmap generation
Infrastructure & Comms 8
The rails that make Mark-1 operational & reachable.
Hostinger VPSProgrammatic VPS provisioning
StripeLive billing, checkout, webhooks
Telegram Bot / MTProtoReal-time messaging channel
OpenAI RealtimeSub-second voice (Marin)
HeyGenLive video avatar
Google WorkspaceGmail, Calendar, Drive, Sheets, Docs
AgentMail / HimalayaEmail inbox (IMAP / SMTP)
LinearIssue / project management

Each row maps to a verified integration skill in ~/.hermes/skills with a working key or auth path. This is the practical difference between "an agent that can search the web" and "an agent plugged into the world's authoritative data feeds" — the latter is what lets Mark-1 show you the real series, the real filing, the real price.

Side By Side

Capability matrix

The same dimensions, both systems. Every cell is a fact, not a slogan. A cross (×) on the Hermes side is not a criticism — a general platform should not ship your business context.

DimensionStock Hermes AgentMark-1
Skills available75 core + 95 optional = 170193 active (89 inherited + 104 net-new)
Persistent contextFlat memory string per turnMemory + Second Brain graph + user profile, applied unprompted
AutonomyCron tool exists; reactive use20 scheduled jobs: sweeps, briefings, SITREPs, backups, watchdogs
Self-correctionmax_turns ceiling onlyGoal-aware stop + mandatory verification
Product surfaceNone — it is infrastructureDataDesk SaaS: chat, voice, billing, admin
Knowledge synthesisAnswers from context windowHybrid retrieval + cited synthesis with honest gaps
Intelligence layerNoneFeeds reasoned against goals; tiered signals; Intel Map
Business awarenessGeneric, anonymousKnows portfolio, properties, family, Nionium goals
Deployment surfaceLocal CLI / gateway25 live project artifacts + multi-VPS tenancy
External data integrationsGeneric web search + browser36 live APIs — finance, science, geodata, comms (source-first)
CodebaseIdentical — both run Hermes v2026.6.5 (HEAD 4219a91df). Mark-1 is upstream + a curated skill/subsystem layer, re-synced via the safe-update skill.
The Past Few Weeks

Recent additions, in order

The concrete work behind the divergence — each a capability that did not exist in Mark-1 before, with the reason it was built.

Most recent

Loop Discipline — goal-aware stop + mandatory verification

Unattended jobs now halt for the right reason and prove their own output against reality before reporting. Closes the "reported done but nothing changed" failure class. Zero new approval prompts for you.

Recent

Weekly Docker cache prune + disk hygiene

A self-verifying cron that reclaimed 253 GB and keeps the production VPS from filling with build cache — with a built-in container-count safety check so nothing is ever damaged.

Recent

Morning Briefing pipeline

7am Tbilisi executive briefing — unread mail, calendar, tasks, docs-to-sign, international + Iran-theater news — delivered as DataDesk cards and Telegram, with sound + desktop alert.

Recent

My Feeds + EU-region YouTube fetch

Aggregated Telegram / RSS / YouTube panel. Solved the EU-region YouTube block via the innerTube WEB-client path — a genuinely hard problem the public API refuses.

Recent

Stripe billing + invite codes live

Real money can flow: live checkout, signed webhooks, $0 invite-code activation. DataDesk became a real commercial product end-to-end.

Ongoing

Multi-tenancy & Hostinger auto-provisioning

Per-tenant isolated VPS model, relay hub, HMAC-signed Brain calls, user-scoped data — the path to selling DataDesk to more than one executive.

Foundation

Second Brain /think synthesis + brain-chat integration

Hybrid retrieval with inline citations and an honest gaps section, wired into the chat with a sources panel — durable knowledge that compounds.

Methodology · How These Numbers Were Counted

Every figure is reproducible

This report does not estimate. All counts were taken directly from the upstream repo checked out on this VPS and Mark-1's live skill directory, by exact set operations.

Correction note: an earlier draft of this page reported "+157%" by comparing Mark-1's 193 against upstream's 75 core skills only — ignoring the 95 optional skills upstream also ships. That figure was withdrawn. The honest comparison is 170 available upstream vs 193 active in Mark-1, of which 104 are genuinely net-new.

Bottom Line

Hermes is becoming more reliable and broader. Mark-1 is becoming more yours.

The upstream agent will keep getting safer and wider — and Mark-1 inherits every one of those gains on each safe update. But the divergence is permanent and intentional: Mark-1 is no longer "a Hermes install." It is an executive operating system that happens to run on Hermes, with 16 capability domains and 104 net-new skills the platform was never meant to carry.

170 upstream available · 193 active · 104 net-new · 16 domains · 36 live APIs
One codebase. One operator. Built upward, deliberately — and counted exactly.