How-to

Who pays for the tokens

Token spend, subscriptions versus API keys, and who pays when you run it for clients.

Type
How-to
Confidence
High
Sources
6 sources
First published
Updated

The common move is to avoid variable API billing by running Hermes on a fixed-price subscription. An SME automation consultant has clients pay their own ChatGPT subscription directly to OpenAI so there are no token costs in the quote, and has the client own and pay for server and model accounts. Marc Bara forces jobs onto his subscriptions while choosing models carefully. @coldemailchris says one shared Codex subscription saves about $800 a month versus seats. Limits bite: an attorney who runs a local business hit a subscription cap during a heavy backfill. Loïc Berthelot says a runaway SEO agent used three times his usage in 15 minutes, so he caps tasks at 90 turns and shares subagent budgets. A Reddit commenter questions how clients who do nothing technical can manage their own API keys, and installers expect security and key ownership to be the most repeated objections. Figures are self-reported.

Key points

  • Put model usage on a fixed subscription, ideally the client’s own, to avoid variable API bills — 3 posts. (an SME automation consultant · Anyone else deploying Hermes into SME… ¶ L388) (Marc Bara · Marc Bara: 15 days with Hermes Agent, 50+… ¶ L94) (a commenter running Hermes on mini PCs · My Hermes setup (Notion knowledge base)… ¶ L270)
  • Share one subscription across a team — 1 post. (@coldemailchris · Our team built an AI agent that builds our… ¶ L12)
  • Cap turns and budgets so a looping agent cannot drain an account — 1 post. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L65)
  • Heavy bursts can hit subscription limits — 1 post. (an attorney who runs a local business · My Hermes setup (Notion knowledge base)… ¶ L97)
  • Client ownership of keys is a selling point and an objection — 2 posts. (an SME automation consultant · Anyone else deploying Hermes into SME… ¶ L17) (a Reddit commenter · I made €2,700 last month installing Hermes… ¶ L502) (unnamed commenter · I made €2,700 last month installing Hermes… ¶ L179)

Tactics

  • Have clients pay for model access on their own existing subscription to avoid variable API billing in your quote. (an SME automation consultant · Anyone else deploying Hermes into SME… ¶ L389)
  • Cap turns and share budget across subagents so a failing API loop cannot drain your account. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L66)
  • Run Hermes on a VPS in Slack on one shared Codex subscription — Hermes runs on a Hostinger VPS 24/7 in Slack with one Codex account underneath instead of a seat per person. (@coldemailchris · Our team built an AI agent that builds our… ¶ L12)
  • Run jobs and sessions on models through existing subscriptions to limit API cost. — Choose carefully which model does which job. (Marc Bara · Marc Bara: 15 days with Hermes Agent, 50+… ¶ L94)

Stories & examples

  • Runaway SEO agent: The author’s first SEO agent read an article library twice and used 3x the author’s usage in 15 minutes. Lesson: The hard cap of 90 turns per task and shared subagent budgets exist for this reason. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L65)
  • Usage limit during heavy backfill: During heavy Q&A backfill and lead gathering, Hermes reported it was limited; a reset in the Codex subscription fixed it. Lesson: Heavy bursts can hit subscription limits. (an attorney who runs a local business · My Hermes setup (Notion knowledge base)… ¶ L97)

Numbers

  • Monthly saving from a shared Codex subscription: 800 USD/month (self-reported, approx.) (@coldemailchris · Our team built an AI agent that builds our… ¶ L13)
  • Usage overrun from the runaway SEO agent: 3 times normal usage in 15 minutes (self-reported) (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L65)
  • Suggested model subscription price: 20 USD/month (self-reported) (a commenter running Hermes on mini PCs · My Hermes setup (Notion knowledge base)… ¶ L270)

Caveats & warnings

  • A Reddit commenter sees a possible contradiction: clients are said to do nothing technical yet handle their own model consumption and API keys. (a Reddit commenter · I made €2,700 last month installing Hermes… ¶ L502)
  • An SME automation consultant says it could work and asks who pays for the model subscription. A commenter selling hosted assistants answers that the user pays, probably through Nous, subscribed at a slight markup, which is where they make their money. (an SME automation consultant · Anyone else deploying Hermes into SME… ¶ L205)

How much to trust this

  • Evidence: 6 posts, 9 independent speakers (repeat posters counted once).

Blueprints that use this

Sources

  1. (01)

    Anyone else deploying Hermes into SME businesses?

    Post

  2. (02)

    Hermes God-Mode Instruction Manual (Dropmagic operator)

    Post

  3. (03)

    Our team built an AI agent that builds our entire TAM (Slack)

    Post

  4. (04)

    My Hermes setup (Notion knowledge base), crons and use cases: attorney-operator on a Raspberry Pi

    Post

  5. (05)

    Marc Bara: 15 days with Hermes Agent, 50+ capabilities

    Post

  6. (06)

    I made €2,700 last month installing Hermes Agent for French companies (post removed; comments remain)

    Post

(By email)

Compounding Insights