This is the recommended starting point for anyone who has never run an agent team. Mavis is not Hermes — they are different products by different companies (MiniMax ships Mavis; Nous Research ships Hermes). Mavis is a $10/month desktop install (Mac or Windows) that bundles an assistant, a coding package, a scheduler, and an adversarial verifier under a single token plan. Once you outgrow Mavis, move to Hermes or a custom Claude Code harness.

The framing the channel uses is sharp: "In Hermes, you have to customize every single config file for specialized agents. Here in Mavis, they already set it up for you." That's the entire pitch. Mavis is the "AI company on easy mode" install at $10/month — agent teams, coding, scheduling, and a built-in adversarial verifier rolled into a single token plan, with the same multi-agent pattern Hermes extends. The Mavis video (30,626 views) is the most-viewed Hermes-adjacent video on the channel and the single best starting point for a beginner.

This article walks through the Mavis framing, the orchestrator + adversarial verifier pattern, the slash-command skills, the scheduled tasks, the new-vs-classic product split, the cost frame, and the explicit "Mavis ≠ Hermes" distinction the syllabus flags. The point is to make the install path obvious: if you've never run an agent team, start with Mavis; once you outgrow Mavis, the next step is Hermes or a custom Claude Code harness.

What you'll learn

  • Mavis ≠ Hermes. Mavis is a MiniMax desktop product (~$10/month); Hermes is the Nous Research CLI/VPS harness. Mavis is the recommended beginner on-ramp; Hermes is the deeper harness for custom workflows.
  • The orchestrator + adversarial verifier pattern: Mavis ships a built-in "devil's advocate" agent that reviews outputs without shared conversation history. Workers produce the work, the verifier checks it from first principles.
  • The slash-command skills and scheduled tasks: type / inside the assistant to invoke reusable skills (PPT generators, presentation makers), or build your own. The scheduler handles daily jobs like SEO keyword monitoring and YouTube competitor analysis.
  • The cost frame: on the same budget, Claude Teams burned through 60% of the weekly quota planning a single trip; the same workload on Mavis "still got enough credits to roll forward."
  • The new-vs-classic product split: use the new Mavis product, not the classic "Cloud Doctor / Pocket" build. The new system "seems to work a lot better."
  • The referral-code discount: 16% off any single subscription tier (verbal); 12% off the Token Plan (pinned comment). These are referral codes, not list prices.

The Mavis framing — and why "multi-agent" is the point

The Mavis video (30,626 views) is the most-viewed Hermes-adjacent video on the channel and the single best starting point for a beginner. The host frames it bluntly: "In Hermes, you have to customize every single config file for specialized agents. Here in Mavis, they already set it up for you." Mavis is positioned as a starter pack — a packaged desktop app at $10/month that gives you the multi-agent pattern without the customisation cost.

What Mavis actually is. A desktop install (Mac or Windows) that bundles an assistant, a coding package, and a scheduler under a single token plan. The host names the assistant explicitly: "This is the Mavis Orchestrator. So, every message you give to even when you start a new task, your Mavis Assistant is the essentially the leader agent or the orchestrator that talks to you." The same orchestrator pattern shows up in Hermes (the Mavis video is the on-ramp, the Hermes videos are the deeper harness).

Price and bundling. The entry tier is $10/month and the creator reports it bundles text, image, and video (Hailuo) under the same token plan — "it's actually combined all rolled into one. So, it's like Hailuo video, it's actually now being combined with their everything… instead of just a token plan, they actually changed it to a full-on token video image text plan everything." On the same budget, the creator reports the same trip-planning workload that burned ~60% of Claude Teams' weekly quota still had credit left on the $10 Mavis plan: "60% of my weekly quota is used… The Chinese ones, you can plan an entire trip, multiple trips, do multiple projects, and you still got enough credits to roll forward."

The new-vs-classic product split. Use the new Mavis product, not the classic "Cloud Doctor / Pocket" build. The creator is explicit that the new system "seems to work a lot better." The classic interface is still around for users with computer use requirements (the new interface lacks those settings, per a viewer comment from @realme72only cited in Course 3 §3.1).

Referral code. The host mentions a 16% referral discount at the end of the video ("remember that we have a referral code down below if you want to sign up for any single plan you get 16% off"); the pinned comment for the same video shows a 12% Token Plan code. These are referral codes, not list prices.

The orchestrator + adversarial verifier pattern

The orchestrator + adversarial verifier pattern is the structural idea that carries from Mavis into Hermes. Mavis ships a built-in "devil's advocate" verifier that reviews outputs without shared conversation history: "the last thing is the verifier… the way it works too with the Mavis verifier is it doesn't have shared conversation history. So, it's not messy. Every question you ask it, it's completely unbiased."

The structural property worth internalising: workers produce, the verifier audits from first principles. The verifier is blind to the worker's conversation history, so the second opinion is genuinely independent. That's the difference between an adversarial verifier and a self-critique loop — the verifier has no incentive to confirm the worker's output.

The host's rule for sub-tasks: "the best performance comes at the early context… having new agents or having a team member of teams actually makes sense here." Spinning up new agents for sub-tasks is actually faster than extending one thread. Lean on the verifier for fact-checking instead of asking the same agent to fact-check its own output. The same pattern shows up in Hermes' multi-agent mode (covered in Course 3 §3.1).

Slash-command skills and scheduled tasks

Two Mavis features make the day-to-day workflow cheap:

  • Slash-command skills. Type / inside the assistant to invoke reusable skills (PPT generators, presentation makers, etc.), or build your own. The pattern is the same as Claude Code's /plan or Hermes' /goal — invoke a skill by name, the assistant loads the skill's instructions into the working set, and the skill's workflow runs.
  • Scheduled tasks. The scheduler handles daily jobs like SEO keyword monitoring and YouTube competitor analysis. The boxmining team uses it for their morning meeting to decide which topics to cover. The cron layer is the same pattern Hermes uses in its Kanban / cron videos.

The onboarding pattern the creator explicitly recommends: tell the Mavis Assistant to "ask me any questions first before you make any changes or any actions" on your first run. The channel flagged this as the right way to keep the orchestrator from going off-script — the assistant will pause to ask for clarification before doing anything destructive. This is the same pattern as the §6.3 dangerous-agent prevention: a /goal with explicit constraints and a definition of done is the cheapest structural fix.

The "Mavis ≠ Hermes" distinction

This is the explicit syllabus flag. Mavis and Hermes are different products by different companies (MiniMax ships Mavis; Nous Research ships Hermes). The Mavis video is the channel's recommended beginner on-ramp and uses the same multi-agent pattern Hermes extends — but Mavis is a packaged desktop app at $10/month, and Hermes is the CLI/VPS harness. Don't try to install "Hermes" from a Mavis tutorial. Don't try to invoke Mavis features from a Hermes config.

The right sequence:

  1. Start with Mavis (this article) — desktop install, $10/month, packaged multi-agent.
  2. Move to Hermes (Course 3) — CLI/VPS harness, custom config files, Kanban + cron + Skill Bundles.
  3. Or move to a custom Claude Code harness (Course 4) — Node CLI, MCP integration, agent teams, env-var swap with Minimax from §5.3.

The decision tree:

  • Beginner / non-technical / "I just want a team of agents on my Mac": Mavis. The host's framing is explicit: "In Hermes, you have to customize every single config file for specialized agents. Here in Mavis, they already set it up for you."
  • Email + Slack + Notion + Google Docs white-collar workflow: bookmark OpenHuman (Course 3 §3.1). The channel's own verdict is "wait" — re-evaluate in 4-8 weeks per the OpenHuman review.
  • Research, multi-step coding, trading, or anything with custom integrations: Hermes. The host's framing: "OpenHuman… is more so for office workers… if you want full customization, you want full autonomy of your own project, then of course, Hermes Agent."
  • Coding specifically, with cheap model routing: Claude Code + Minimax from Course 4 and §5.3 of this course.

The cost frame — Mavis vs Claude Teams

The cost frame the channel uses is direct: on the same budget, Claude Teams burned through 60% of the weekly quota planning a single trip; the same workload on Mavis "still got enough credits to roll forward." That's the operational case for Mavis as the beginner on-ramp: a $10/month flat-fee plan that gives you multi-agent orchestration, scheduled tasks, slash-command skills, and an adversarial verifier, all bundled under a single token plan that includes text, image, and video.

The token plan resets on the same 5-hour rolling counter as the Minimax token plan from §5.3. That's not a coincidence — Mavis is the MiniMax desktop product, and the Minimax token plan is the underlying engine. When you subscribe to Mavis, you're getting the Minimax executor slot from §5.1 wrapped in a packaged multi-agent UI.

Audience signal

The Mavis video's top substantive viewer comment is from @RyanDavisEdwardjr (2026-05-20, 1 like): "Mavis came out like 3 weeks ago and you are the first people covering it. Thank you. There is a huge gap in market people wanting to learn cheap AI orchestration and tools." A second viewer (@realme72only, 2026-05-20, 0 likes) confirms the cost / Apple Mail caveat: "the new token-based plan is way more budget-friendly than the old credit system—I haven't even hit my limits yet… the new interface is super clean but lacks 'computer use' settings. You'll need to use the classic interface for that, though I still had to use AppleScript to get it to control Apple Mail properly." The BoxminingAI host also pinned a 12% Token Plan referral link and the Agent Desktop download.

Try it yourself

  1. Subscribe to Mavis at the $10/month entry tier. Use the referral code from the pinned comment (12% off the Token Plan) or the verbal 16% off any single tier.
  2. Install the desktop app on Mac or Windows. The agent is bundled — no separate config files, no separate model wire-up.
  3. On your first run, ask the assistant to "ask me any questions first before you make any changes or any actions." This is the onboarding pattern the creator explicitly recommends. It keeps the orchestrator from going off-script on its first session.
  4. Try the orchestrator + adversarial verifier. Ask the same question to the worker agent and the verifier, then compare. The verifier is blind to the worker's conversation history, so the second opinion is genuinely independent.
  5. Build a slash-command skill. Type / inside the assistant to see the bundled skills (PPT generators, presentation makers). Pick one and customise it for your workflow. The pattern transfers to Hermes' Skill Bundles and Claude Code's /plan primitive.
  6. Schedule a daily task. Pick something routine — SEO keyword monitoring, YouTube competitor analysis, a daily news brief. The scheduler uses the same cron pattern as the OpenClaw / Hermes cron layer.
  7. Spin up a fresh agent per sub-task instead of extending one long thread. The channel's data: "the best performance comes at the early context," so new agents beat thread continuation.
  8. When you outgrow Mavis, move to Hermes or Claude Code. Mavis is positioned as a starter pack, not a replacement. Once the bundled multi-agent UI starts to constrain you, the next step is the Hermes Kanban + Skill Bundles pattern from Course 3 §3.3 and §3.5, or a custom Claude Code harness from Course 4.

Common pitfalls

  • Confusing Mavis and Hermes. They are different products by different companies. Mavis is a MiniMax desktop product; Hermes is a Nous Research CLI/VPS harness. The Mavis video is the channel's recommended beginner on-ramp and uses the same multi-agent pattern Hermes extends — but the products are not interchangeable. Don't try to install "Hermes" from a Mavis tutorial. Don't try to invoke Mavis features from a Hermes config.
  • Letting the orchestrator go off-script on the first run. The onboarding pattern is to ask the assistant to "ask me any questions first before you make any changes or any actions." Skip this step and the first session may produce a side effect you didn't intend.
  • Asking the same agent to fact-check its own output. Use the adversarial verifier. The verifier is blind to the worker's conversation history, which is the property that makes the second opinion independent. Asking the worker to critique itself produces a confirmation bias, not a check.
  • Extending one long thread instead of spinning up fresh agents. The channel's data: "the best performance comes at the early context." Continuing a stale thread will look worse than starting a new sub-agent.
  • Treating the $10/month plan as a fixed budget. The token plan resets on a 5-hour rolling counter, but the Mavis bundle is generous. Heavy cron work overnight is fine. Heavy interactive work all day will burn through the cap. Plan around the window.
  • Using the classic "Cloud Doctor / Pocket" build. Use the new Mavis product. The creator is explicit that the new system "seems to work a lot better." The classic interface is for computer use workflows that the new UI doesn't yet support.
  • Trying to get computer use working in the new interface. Per a viewer comment, the new interface lacks computer use settings. If you need computer use, use the classic interface, and even then you'll need AppleScript for Apple Mail.
  • Stopping at Mavis. The product is positioned as a starter pack, not a replacement. Once the bundled multi-agent UI starts to constrain you, the next step is Hermes or a custom Claude Code harness.
  • Paying Claude Teams prices for executor work on a beginner workflow. The cost frame is direct: Claude Teams burned 60% of the weekly quota planning a single trip; the same workload on Mavis had credit left. If you're a beginner, route the executor work to Mavis before you route it to Opus.

Sources

  • Minimax Mavis: The BEST Multi-Agent Platform for Beginners — 30,626 views · video_id: 86UIZVWkvF8 · watch
  • Hermes vs OpenClaw: Why Everyone Is Migrating — 6,116 views · video_id: 2NbfOOD2i1E · watch
  • Top AI Models for Hermes Agent (Tier List) — 8,107 views · video_id: Af7Fg1m7hRw · watch
  • Minimax M2.7 is INSANELY GOOD! (Full Review) — 31,049 views · video_id: --uxieT5J9Y · watch
  • Is Minimax the Best AI Model for OpenClaw? — 3,219 views · video_id: 258R3kzDRAQ · watch
  • Claude Code + Minimax 2.7: Unlimited AI Coding on a Budget — 6,532 views · video_id: dURSH_Fwu6s · watch
  • MiniMax M3 is HERE! (Real Tests and Review) — 4,398 views · video_id: -Qf3bvFTIzY · watch
  • Minimax M3 Review: Is This Cheap AI Model Actually Worth It? — 856 views · video_id: hTkxebQdtH8 · watch
  • Supabase query — SELECT video_id, title, views, summary_content, summary_key_takeaways FROM public.videos WHERE video_id = ANY(ARRAY['86UIZVWkvF8','2NbfOOD2i1E','Af7Fg1m7hRw','--uxieT5J9Y','258R3kzDRAQ','dURSH_Fwu6s','-Qf3bvFTIzY','hTkxebQdtH8']); against project ttxdssgydwyurwwnjogq.
  • Cross-references to the syllabus sections this article teaches into: Course 1: Picking Your Agent Harness (the agent platform Mavis sits on top of), Course 2: AI Models §2.3 (Minimax — the engine under Mavis), Course 3: Hermes Agent (the multi-agent platform downstream of Mavis), Course 4: Claude Code & AI Coding (the harness the §5.3 env-var swap plugs into), Course 6: Agent Memory & Troubleshooting §6.2 (the dumb-zone diagnostics) and §6.5 (the /goal primitive that pairs with the Mavis orchestrator's "ask me first" pattern).

NOTE on pricing, version numbers, and roadmap claims: the 1/16th cost ratio ($5 vs $0.30 per million input tokens), the M2.7 architecture (230B MoE / 10B active per token), the M2.5 76.8% BFCL score, the M3 SWE Bench Pro 59% score, the M3 MSA 9.7× pre-fill / 15.6× decode speedups, the M3 API pricing ($0.30 / $1.20 per million input/output), the 4,500/5h Plus and 15,000/5h Max token-plan request limits, the 100 prompts / 5-hour starter plan, the $10/month Mavis entry tier, the 16% referral discount (verbal) and 12% Token Plan code (pinned comment), the 3/day Max and 5/day Ultra M3 video generation limits, the 200K token-plan context cap, the 300-line dumb-zone threshold for soul.md, the 15-30 line compressed target, the 7-10 skills-per-agent cap, the 40% context threshold on MiniMax-class models, the Mavis "new vs classic" product split, and the Mavis ≠ Hermes distinction are all drawn from the source videos cited above. These are time-stamped claims — re-check the official Minimax and MiniMax Code documentation if you read this course after a new release.

Aggregate Sources across all 5 articles. The 8 videos cited across this course, in order of view count: Minimax M2.7 is INSANELY GOOD! (Full Review) — 31,049 views · video_id: --uxieT5J9Y · watch · §5.1, §5.2, §5.4 cross-ref, §5.5 cross-ref. Minimax Mavis: The BEST Multi-Agent Platform for Beginners — 30,626 views · video_id: 86UIZVWkvF8 · watch · §5.5. Top AI Models for Hermes Agent (Tier List) — 8,107 views · video_id: Af7Fg1m7hRw · watch · §5.1, §5.2, §5.5 cross-ref. Claude Code + Minimax 2.7: Unlimited AI Coding on a Budget — 6,532 views · video_id: dURSH_Fwu6s · watch · §5.3, §5.5 cross-ref. Hermes vs OpenClaw: Why Everyone Is Migrating — 6,116 views · video_id: 2NbfOOD2i1E · watch · §5.5 cross-ref. MiniMax M3 is HERE! (Real Tests and Review) — 4,398 views · video_id: -Qf3bvFTIzY · watch · §5.1, §5.5 cross-ref. Is Minimax the Best AI Model for OpenClaw? — 3,219 views · video_id: 258R3kzDRAQ · watch · §5.1, §5.2, §5.3, §5.4, §5.5 cross-ref. 5 Must Know TIPS for OpenClaw — 3,473 views · video_id: -PT46iH03RQ · watch · §5.4 cross-ref. How to Build Your OpenClaw AI Agent the RIGHT Way — 2,690 views · video_id: Zkw8jIDzspc · watch · §5.4. My OpenClaw is STUPID (Here's how to Fix It) — 1,535 views · video_id: 9lcn8ZmqyJ0 · watch · §5.4. Why Your AI Agent Suddenly Gets Stupid And How to Fix It — 767 views · video_id: pMgUaqXTge4 · watch · §5.4. Minimax M3 Review: Is This Cheap AI Model Actually Worth It? — 856 views · video_id: hTkxebQdtH8 · watch · §5.1, §5.2, §5.5 cross-ref. Why MINIMAX M 2.7 WINS! Parallel Subagents and Self Auditing — 669 views · video_id: ocjfBoM_eTM · watch · §5.2. MiniMax M2.7's Best Feature Nobody's Using (Multi-Agent Teams) — video_id: Ttb_Tw6-YBA · watch · §5.2.