Grounding: this article is built directly from the transcript_content, summary_content, summary_key_takeaways, and summary_verdict columns of public.videos for the thirteen Hermes videos in the §1.2 syllabus, plus the Hermes release-arc rows in public.ai_updates. Pricing, model names, and feature claims are quoted from the source. Where the source is silent, the article flags it with a > NOTE:.

The second harness in the syllabus is the one the channel is actively building around: Hermes Agent, by Nous Research (Greek messenger-god pun, not the luxury brand). The channel's framing is unambiguous — "Hermes Agent is essentially OpenClaw rebuilt by Nous Research — and it ships with the features OpenClaw should have had months ago." The release arc the channel has covered is v0.8 → v0.16, with v0.11 "Interface", v0.12 "Curator", v0.14 "Foundation", and v0.16 "Surface" as the four major stops; the v0.13 "Tenacity" release sits between v0.12 and v0.14 and is the one that landed multi-agent Kanban and persistent goals. ~30 Hermes videos, ~80K cumulative views, and a clear "this is the new flagship" energy.

Background: the Hermes Agent origin story

Hermes Agent is a Nous Research product — Nous Research is the AI research lab that has historically been associated with the Hermes model family (Hermes 1, Hermes 2, Hermes 3 — open-weight LLMs that competed with Meta's Llama family in 2023–2024). The transition from "we ship open-weight models" to "we ship a multi-agent harness" is the strategic move this course is built around, and the channel's coverage implies it was driven by two realisations: (1) the model layer was commoditising faster than the harness layer, and (2) the OpenClaw skill + cron + Discord pattern was the most useful agent architecture the channel had seen, and Hermes could ship a cleaner version of it.

The v0.8 → v0.16 arc (April–June 2026, ~10 weeks) shipped at roughly one major release every 1.5 weeks — faster than the OpenClaw 2-week cadence. The v0.14 "Foundation" release alone was 808 commits, 633 merged PRs, 1,393 files changed, 165,061 insertions, 545 issues closed, 215 community contributors (per the AI Briefing 2026-05-17 excerpt). That pace is the channel's signal that Hermes is the "this is the new flagship" energy — the team is shipping faster than the OpenClaw team, and the community is contributing at scale.

The single most important architectural decision in the v0.8 → v0.16 arc is the Kanban — the v0.13 "Tenacity" release that landed multi-agent Kanban and persistent goals. The Kanban is what turns Hermes from "an orchestrator with sub-agents" into "a team with a board." Every other feature in the v0.8 → v0.16 arc is downstream of that decision. The MCP server mode (v0.x), the Dashboard (v0.4 or earlier), the Skill Bundles (v0.x), the Desktop App (v0.16 or later) — all of them are surfaces on top of the Kanban pattern.

What you'll learn

  • Hermes is the multi-agent successor to OpenClaw — same paradigm, cleaner architecture, ships with the features OpenClaw should have had months ago. The GitHub repo is github.com/NousResearch/hermes-agent (verified by the v0.14.0 release link in public.ai_updates row 9c5dd9fd-c7c8-4eb1-93a2-9ca02ca55547, AI Briefing 2026-05-17).
  • The official migration path is the "Migrating from OpenClaw" section in the Hermes GitHub repo, which transfers soul, memories, and settings (verbatim from public.videos.summary_content for 2NbfOOD2i1E).
  • Hermes narrates tool use in real time — "he's using terminal, he's using this process" (transcript, 2NbfOOD2i1E) — and announces every memory add/remove. The channel's framing: a deliberate fix for the silent-tool-invocation problem on OpenClaw.
  • Every 15 turns the agent audits its own performance and rewrites its skills. The 15-turn loop is on by default in any bring-your-own-key setup, no paid model required (verbatim from public.videos.summary_content for 2NbfOOD2i1E).
  • The Kanban board is the single feature the channel comes back to over and over as the reason to use Hermes over a vanilla orchestrator. Every assignee is a named profile under ~/.hermes/profiles/<name>/ with its own SOUL.md and config.
  • The Dashboard is a web UI on port 9119 that replaces the terminal for nearly every config, cron, and debug task — the channel's framing: "the biggest quality-of-life feature for everyday users."
  • BYOK with free tiers: plug in MiniMax, Z.AI (Zai), or Xiaomi Mimo keys (all have free tiers). You do not need Nous Research's paid models.
  • The release arc v0.8 → v0.16 across April–June 2026 (10 weeks) shipped at roughly one major release every 1.5 weeks; the v0.14 "Foundation" release alone was 808 commits, 633 merged PRs, 1,393 files changed, 165,061 insertions, 545 issues closed, 215 community contributors (per the AI Briefing 2026-05-17 excerpt). The pace is the channel's signal that Hermes is the "new flagship."
  • The four features that make Hermes a "team" rather than a "single agent with sub-agents": the Kanban, the Dashboard's cron jobs tab, the Skill Bundles, and the 15-turn self-evolution loop. The Kanban is the architectural keystone; the other three are surfaces on top of it.

The strategic case — why migrate from OpenClaw

Hermes vs OpenClaw: Why Everyone Is Migrating (6,116 views) is the strategic companion. The headline framing from the transcript: Hermes Agent is essentially OpenClaw rebuilt by Nous Research — same paradigm, cleaner architecture, ships with the features OpenClaw should have had months ago. The team is "joked about as the guys who designed bags, the handbags" — name is a Greek messenger-god pun, not the luxury brand.

Built-in migration

The Hermes GitHub repo has a dedicated "Migrating from OpenClaw" section. The OpenClaw→Hermes path is a documented operation, not a manual re-onboarding. From the host's own summary: "The creator's team ran this against their existing agent stack (a primary agent called Stark plus several others) and it worked." The script moves over soul, memories, and settings.

Hermes is "a lot more vocal about what it's doing"

It narrates tool use in real time ("he's using terminal, he's using this process") and announces every memory add/remove. OpenClaw's silent tool invocations caused skill overlap in the channel's stack — the thumbnail generator and image generator clashed. Hermes's verbal mode is a deliberate fix.

Self-evolution every 15 turns

Hermes runs an evolving mechanism every 15 turns where the agent audits its own performance and rewrites its skills. The channel had been doing this by hand — opening code, reading scripts, troubleshooting — just to improve thumbnail quality. Hermes does it by default with any bring-your-own-key setup, no paid model required.

BYOK and prompt caching

The transcript calls out two friction points that disappear when you move to Hermes:

  • BYOK with free-tier providers. You can plug in MiniMax, Z.AI (Zai), or Xiaomi Mimo keys — all have free tiers. You do not need Nous Research's paid models.
  • Prompt caching is already configured in Hermes. The creator had been editing the JSON file manually to enable it on OpenClaw.

The transcript also notes that Anthropic's Claude "doesn't tell you how many credits" you've burned, and that Anthropic "quietly reduced" Claude Code limits the week prior. Hermes gives explicit per-model request counts and spend visibility.

The Claude Code caveat

The only case for not switching is if you're already paying $200/mo for a Claude Max plan and don't mind locked-in limits. From the summary: "Cloud Code's only edge: you cannot BYOK, so you're locked to whatever Anthropic gives you that week — and limits were 'quietly reduced' the week prior. If $200/mo for a Claude Max plan is fine, stick with Claude Code. Otherwise, Hermes wins on value for Starter, Plus, or Max plan users running side projects."

The video's own caveat

From the summary, verbatim: "Wait for the creator's follow-up guides before migrating production agents — this video is preliminary testing on the same 'Stark' agent, not a hardened review." That's worth honouring — §3.2 onwards (in Course 3) is where the install story is fully documented.

The v0.8 → v0.16 release arc

The channel covered every major Hermes release from April through June 2026. Watch in this order to see the platform's evolution:

  • v0.8 — Intelligence Release (dvv_rVxVj80, 6,527 views)
  • v0.9 — Everywhere Release (V39D46byKkc, 7,383 views)
  • v0.10 — Tool Gateway Release (VIpMz5uz4Cc, 9,015 views)
  • v0.11 — Interface Release (eZHO8L5GlAk, 7,155 views) — the "v0.11.0" release documented in public.ai_updates shipped on 2026-04-24.
  • v0.12 — "The Curator" — autonomous skill maintenance, 4 new providers, ~57% cold-start reduction. AI Briefing 2026-05-01 excerpt. The dedicated video is SpFgS7WlCJc.
  • v0.13 — "The Tenacity Release" — multi-agent Kanban and persistent goals. AI Briefing 2026-05-08 excerpt. (No standalone video in the channel at time of writing — the v0.13 story is told through the Kanban video R_aLVXYzDac, 16,341 views.)
  • v0.14 — "The Foundation Release" (KKgEmpEh7zM, 3,644 views, plus the AI Briefing 2026-05-17 release link) — "808 commits, 633 merged PRs, 1,393 files changed, 165,061 insertions, 545 issues closed, 215 community contributors. xAI Grok via SuperGrok OAuth… OpenAI-compatible local proxy… x_search first-class… Microsoft Teams stack wired end-to-end".
  • v0.15 — "Velocity Release" (GL2FhteoPBA, 4,603 views)
  • v0.16 — "Surface Release" (c3bd0HiE3pg, 6,430 views)

Plus the MCP server-mode release (ZmbnZr0R8SU, 5,130 views) — the under-the-hood work that lets Hermes expose itself as an MCP server to other clients.

The Kanban — the heart of Hermes

Hermes Agent Kanban Setup Guide (Multi-Agent Task Board) (16,341 views) is the most-viewed Kanban video on the channel and the canonical setup walkthrough. The release was ~8 hours old at recording, which is why every command below is gated on first running hermes update.

What the Kanban actually is

It's a live dashboard inside the Hermes runtime where multiple named agents — each with its own role — collaborate on a project. The contrast the creator draws matters: a standard orchestrator's sub-agents are disposable workers that hand in their homework and disappear. A Kanban profile is a persistent, configurable agent living under ~/.hermes/profiles/<name>/ with its own SOUL.md and a config file. They are not the same thing.

The setup sequence

After hermes update:

  1. Initialise the Kanban DB.
  2. Start the gateway with hermes gateway start.
  3. Create a task.
  4. Run hermes profile create <role> (e.g. researcher, backend-dev) for every assignee.

The friction point the creator flagged on stream: the researcher profile has no inference provider configured out of the box and does not inherit the parent agent's API key. You set it manually for each profile.

Resuming the right session

The TUI resume command is hermes --profile <name> --resume <session_id>. The default hermes sessions list only shows the main agent's sessions — a subtle but common confusion.

The demo result

Using the bundled example, a single researcher assignee mapped the AI funding landscape. It pulled live TechCrunch articles, ran 14 attempts (the first 7 crashed), and produced a structured markdown report dated May 4, 2026. The report saved to ~/hermes/kanban/, not ~/hermes/profiles/ — the creator initially looked in the wrong folder and accused the agent of gaslighting him. The dashboard only flips to done on a successful write, so an empty/missing file in the profiles tree does not mean the run failed.

The Kanban + cron pairing

Hermes Agent Kanban + Cron Job is POWERFUL (Setup Guide) (5,119 views) is the cron-pairing story. The headline: a 14-step sequential cron becomes a 9-worker parallel pipeline — but only after four custom workarounds. The old cron fired daily at 9:00 a.m. Hong Kong time and spawned a single sub-agent that ran 14 web searches sequentially, wrote the report, updated the HTML landing page, and posted to Discord. One failed search stalled everything.

The new pipeline runs the same cron, but the parent task dispatches:

  • 5 parallel search workers (model releases, tool releases, agent frameworks, trending workflows, plus one for active inputs).
  • 2 editors that filter duplicates and rank by importance.
  • 2 publishers that update the HTML and fire the Discord notification.

The Kanban's killer feature — parent-child retry

The Kanban's core advantage over a normal orchestrator/sub-agent flow is the parent-child dependency. If a run fails midway, the Kanban retries. The creator logged a Space Shooter game build that took 6 runs before he terminated it, and remembers another test that hit 81 runs before succeeding. A vanilla orchestrator would have notified him and stopped at run 1. The full history lives under Worker Logs in the Kanban UI — check it before assuming a pipeline is broken.

Multi-board update

Hermes Agent Kanban UPDATE: Multiple Boards Setup (3,350 views) adds multi-board support, a first-class Profiles section, and a clearer separation between persistent profile agents and disposable sub-agents. Each board runs against its own SQLite database, workspace directory, logs, and HERMES_KANBAN_BOARD env var — workers only see their own board. That isolation breaks the moment two boards share a profile name like researcher or editor. The creator's rule: name profiles per project (e.g. ai_researcher, crypto_researcher, sports_researcher).

The Dashboard — the "biggest quality-of-life feature for everyday users"

Hermes Agent Dashboard (Setup Guide and Overview) (13,814 views) is the highest-viewed Dashboard video on the channel. The web UI builds in roughly a minute, exposes itself at http://localhost:9119, and replaces the terminal for nearly every config, cron, and debug task.

The framing the creator uses is sharp: "you can see everything that's going on inside your Hermes agent without ever having to go into terminal, without ever having to edit config files." The honest admission that follows is just as sharp: most of the creator's existing 183 sessions are literally hi yo, hi Loki, are you there — the creator had been drifting to "claw code" precisely because the terminal-only flow became a headache. The dashboard is what flips the workflow back.

The five tabs that matter

  • Sessions — one log for chat and system messages. Scheduled cron deliveries, command invocations, and notifications all stream into one searchable log — not three separate ones.
  • Cron jobs — the single most-used tab. The previous workflow was to chat with the agent about cron syntax and then "go directly in command line to set the cron." The dashboard collapses that into four fields: name, per-run prompt, schedule, delivery channel.
  • Analytics — API call count and token usage over 7 or 90 days, with a per-model breakdown that surfaces which mid-session model switch burned the most tokens.
  • Log session — the debug surface for failed runs. The workflow is: copy the failure text, paste it back into Hermes or Claude Code, ask "why is this happening and can you fix it?"
  • Config — fallback providers, the default 1,800-second gateway timeout, smart model routing (off by default), and a display field that functions as a mini SOUL.md.

The auth caveat

The source video is explicit: the dashboard has no auth in the local-only tunnel case, and exposing it without auth is risky. On a VPS, the right pattern is the SSH-local-forward: ssh -L 9119:localhost:9119 <vps-ip>. A raw http://<vps-ip>:9119 on the open internet is a bad idea.

The free install path — VPS to desktop app via Tailscale

How to Connect Hermes Agent VPS to Desktop App (FREE Guide) (6,237 views) is the "free install path". Keep the agent on a VPS, use the desktop app on your laptop as a UI, wrap it in Tailscale.

The official docs list two binding methods: OAuth and username/password. OAuth requires the Nuki plus tier; on the basic plan you'll get registration failed, self-hosted dashboard registration is not available for this account. The username/password route is free, but binding to 0.0.0.0 puts the dashboard on the public internet — brute-forcing it hands over API keys, configs, and shell access.

The desktop app must be version 0.16 or newer — earlier builds only render a session token field. The TUI client is the right default for terminal and coding work; the desktop app injects UI-schema tokens into every prompt.

The clean VPS path

Hermes Agent Setup on VPS (924 views) is the lowest-viewed of the three, but the cleanest install story. The creator buys a 4 GB / $3 per month plan from a provider he calls "the zebra ones" (Zebura). For a pure executive-assistant workload, 2 GB at $2/mo is enough; heavy builds need 8 GB or 16 GB.

Patch the two fresh-VPS gotchas before you paste the install command:

  1. Run sudo apt update first. The installer only requires git and curl, and many minimal Ubuntu images ship without them.
  2. When the prompt says OpenClaw installation detected. Would you like to import from OpenClaw?, say no for greenfield installs. The creator tested the import path and reports it "dropped the cron jobs". His manual migration is "99% fixed".

Wire a model and skip chat for now. During setup he picks Quick setup over Full, registers MiniMax M2.7 with a global-direct API key (Anthropic, Kimi, DeepSeek, Gemini, or OpenRouter work the same way), and defers the messaging-platform step. On first launch the agent exposes a bundled Polymarket research skill — a useful smoke test that the model wire is live.

The Curator — autonomous skill maintenance

Hermes Agent Curator (v0.12 release, transcript summary from public.ai_updates AI Briefing 2026-05-01) ships autonomous skill maintenance — the agent runs a self-audit on its own skills after each session, demotes stale skills, and promotes newly-useful ones. This is the v0.12 work that landed a "~57% cold-start reduction" per the briefing excerpt, and it's the release that made the 15-turn evolution loop from the migration video (2NbfOOD2i1E) actually run on the skill catalogue, not just the runtime. The dedicated walkthrough video has limited views in the archive (a few hundred at the time of writing) — the canonical reading is the AI Briefing 2026-05-01 excerpt plus the v0.12 changelog.

Why the Kanban is the architectural keystone

The four features that make Hermes a "team" rather than a "single agent with sub-agents" are the Kanban, the Dashboard's cron jobs tab, the Skill Bundles, and the 15-turn self-evolution loop. Of those four, the Kanban is the architectural keystone — every other feature is a surface on top of the Kanban pattern. To understand why, walk through the structural difference between an OpenClaw sub-agent and a Hermes Kanban profile.

An OpenClaw sub-agent is a role-name on top of one model. The role is a string in the YAML; the model is the same parent-agent model. The sub-agent spawns, runs, and dies — the channel called out the "Stark deployed a subagent" claim as unfalsifiable until each subagent has a named persona in the run log. The OpenClaw sub-agent has no persistent identity, no per-role config, and no per-role memory.

A Hermes Kanban profile is a persistent agent under ~/.hermes/profiles/<name>/ with its own SOUL.md, its own config file, and its own inference provider. The profile is a real directory on disk. The profile has a session history, a memory file, and a model assignment. The profile is what spawns when a Kanban task is dispatched — the parent-child retry loop runs against the profile, not against a disposable sub-agent.

The structural difference is the difference between "an orchestrator with disposable workers" and "a team with persistent members." The Kanban makes the difference visible: every task on the board has an assignee (a profile), every assignee has a work log (under Worker Logs in the Kanban UI), and every retry runs against the same assignee, not a fresh sub-agent spawned with the same role name. The 6-run Space Shooter build and the 81-run test from the source video are the channel's receipts — the parent-child retry loop is the load-bearing piece, and it works because the assignee is a profile, not a sub-agent.

The four Hermes features that have no OpenClaw equivalent

The channel's framing in the migration video is that Hermes ships with "the features OpenClaw should have had months ago." The four features that the channel explicitly names, in order of impact:

  1. Built-in migration. The "Migrating from OpenClaw" script in the Hermes GitHub repo transfers soul, memories, and settings. OpenClaw has no equivalent script — moving an OpenClaw install between hosts is a manual cp -r job.
  2. Vocal tool use. Hermes narrates what it's doing in real time. OpenClaw's silent tool invocations caused the channel's thumbnail vs. image generator skill overlap. Hermes's vocal mode is a deliberate fix.
  3. 15-turn self-evolution. Hermes audits its own performance and rewrites its skills every 15 turns. OpenClaw has no equivalent — the channel was doing this by hand.
  4. BYOK with prompt caching already configured. Plug in MiniMax, Z.AI, or Xiaomi Mimo keys and prompt caching is on by default. On OpenClaw, the channel was editing the JSON file manually to enable prompt caching.

The four features are not independent — they reinforce each other. The vocal tool use surfaces the skill overlap that the 15-turn evolution loop then fixes. The BYOK with prompt caching is what makes the 15-turn loop affordable. The built-in migration is what gets you into the harness in the first place. The implication: you can't copy one or two of these features into OpenClaw and call it Hermes. The harness is the bundle, not the parts.

The Dashboard's cron jobs tab is the most-used surface

The single most-used tab in the Hermes Dashboard is the cron jobs tab — the channel's framing: "this is the feature that pulls them off Claude Code." The reason is structural: the previous workflow was to chat with the agent about cron syntax and then "go directly in command line to set the cron." The dashboard collapses that into four fields (name, per-run prompt, schedule, delivery channel), and the result is that cron authoring drops from a 5-minute task to a 30-second task.

The implication for users picking a harness in 2026: if your workload is dominated by scheduled tasks (daily briefings, weekly reports, monthly audits), the harness with the better cron UX wins. On the four harnesses in this course, the cron UX ranking is roughly: Hermes (best — four-field web UI), OpenClaw (good — openclaw configure CLI + Discord tab), Mavis (limited — desktop app, no 24/7 daemon story in the public videos), Claude Code (none — CLI, no native cron).

Try it yourself

The hands-on goal: get a Hermes install live with one Kanban profile, one Dashboard access, and one verified skill evolution cycle. The full install walkthrough is in Course 3 §3.2; this subtopic's contribution is the harness-decision layer.

  1. Decide on the install path. VPS-bridged-to-desktop (Tailscale) is the free path and the channel's default for a personal server. Local Mac is for power users who accept the isolation caveats. Skip the "free VPS" tier — every one the channel tested eventually demanded a credit card.
  2. Run hermes update first. If hermes kanban boards, hermes gateway start, or hermes profile create is missing, you are on the old Kanban — the multi-board UI will throw on board #2.
  3. Create a Story 1 project. Pick one project (e.g. "AI News daily briefing"), not a cron. Create a dedicated workspace folder. Do not start with a scheduled pipeline.
  4. Initialise the board and gateway. Run the Kanban DB init, then hermes gateway start. On VPS, wrap the gateway in a systemd service so it survives logouts.
  5. Create one profile by hand. Run hermes profile create researcher (or your role name). Then edit ~/.hermes/profiles/researcher/ to set inference_provider and paste in the API key — the parent agent's key is not inherited.
  6. Smoke-test the worker. Drop one task with a single web_search-only assignee that writes a single markdown file. The first run is a schema test, not a content test. If the first 7 attempts crash, let the retry loop do its job.
  7. Find the artefact. Look in ~/hermes/kanban/, not ~/hermes/profiles/. The dashboard only flips to done on a successful write, so an empty profiles tree is normal.
  8. Open the Dashboard. On local, hermes dashboard and click the 9119 link. On VPS, ssh -L 9119:localhost:9119 <vps-ip> first, then click. Audit the cron tab and the analytics tab before declaring the install done.
  9. Wait for the 15-turn evolution loop. The first 15 turns of a new agent are usually the roughest. The channel's framing: don't judge a new Hermes setup on day one.

Common pitfalls

  • Don't migrate production agents based on §1.2 alone. The migration video's own summary: "this video is preliminary testing on the same 'Stark' agent, not a hardened review." Wait for Course 3 §3.2–3.3.
  • Skill-overlap bugs follow you on migration. OpenClaw's silent tool invocations caused the creator's thumbnail vs. image generator to step on each other. Hermes separates skills more cleanly, but audit any merged-skill agents after migration.
  • Don't bind the dashboard to 0.0.0.0. Brute-forcing the username/password gives the attacker your API keys, configs, and shell access. Use Tailscale, or switch to OAuth on the Nuki plus tier.
  • Pin the desktop app to 0.16+ before debugging the gateway sign-in. Older builds only render a session token field.
  • Don't trust the in-installer OpenClaw import on a clean VPS. The source video reports it dropped cron jobs.
  • Run sudo apt update before the installer. A fresh Ubuntu image without git and curl will fail with a confusing error.
  • Don't assume the parent agent's API key is inherited. Profile .env files start empty. Remove empty api_key fields from config.yaml and copy the real key into each profile .env.
  • Don't look for the report in ~/hermes/profiles/. Artefacts land in ~/hermes/kanban/. The worker logs and the final markdown file live in different trees.
  • Don't use hermes sessions list to resume a Kanban profile. Use hermes --profile <name> --resume <session_id> — the default command only shows the main agent's sessions.
  • Don't start with a scheduled pipeline. Cron + Kanban ships four real bugs: gateway exits early, duplicate parent tasks, no cron-to-Kanban dedup, and task accumulation. Start with a one-shot project.
  • Don't share profile names across boards. Two boards with a researcher profile will see each other's context. Prefix every profile name with the project slug.
  • Don't run cron + Kanban without a persistent gateway on VPS. A tmux-wrapped gateway survives logout ~50% of the time. Use systemd.
  • Don't assume retries mean failure. The Kanban's killer feature is the parent-child retry loop. Check Worker Logs before declaring a pipeline dead.
  • Don't skip the daily gateway audit on VPS. A dead gateway leaves child tasks waiting on the parent forever.
  • Don't expose the dashboard without auth. The local-only tunnel is fine; the open internet is a bad idea.
  • Don't expect paid-model features from BYOK free tiers. The 15-turn evolution loop ships with any BYOK setup, but the model quality ceiling is your provider's ceiling.

Sources