Skip to content

Add Hermes Agent + Token Station tutorial (en/zh/ja/ko) - #41

Merged
juntao merged 1 commit into
second-state:mainfrom
ethanwestfall:content/hermes-agent-token-station
Sep 30, 2026
Merged

juntao merged 1 commit into
second-state:mainfrom
ethanwestfall:content/hermes-agent-token-station

Conversation

@ethanwestfall

Copy link
Copy Markdown
Contributor

Summary

  • Walks through routing Hermes Agent (Nous Research's open-source autonomous agent) through Token Station via a custom OpenAI-compatible endpoint, using OpenAI's GPT-5.6 family (Sol, Terra, Luna)
  • Centerpiece demo: real model delegation confirmed working end to end. openai/gpt-5.6-sol plans and delegates a bounded subtask to openai/gpt-5.6-luna via delegate_task, and both models show up billed separately on Token Station's Usage page in the same session, interleaved minute to minute
  • Also confirms real tool-use (a file actually written and executed, not just described) for openai/gpt-5.6-sol, the same trap that hit the earlier Cursor+OpenAI article before its fix landed
  • Explicitly contrasts with Cursor's subagent model: field, which our earlier Claude Sonnet 5 + Haiku article documented as valid, documented syntax that turns out to be a no-op, subagents always inherit the parent's model there. Hermes's delegation.model/delegation.provider genuinely routes
  • Ships in all four languages (en/zh/ja/ko) plus a dedicated cover image
  • Adds sitemap.xml entries for all four language URLs

An adversarial review agent verified every factual claim, cross-checked pricing against gpt-5-6-token-station.md, verified the Cursor no-op citation against its actual source article, confirmed all four media assets exist with no PII in the screenshot, and confirmed house-style compliance. Two rounds of numbers-accuracy fixes went in before this PR was opened: the stale "$1 in signup credit" claim (removed site-wide as of this promo change, kept the 100% first-top-up match) and a platform-wide "300+/25+ models" catalog-size claim (cut entirely after the author's own account dashboard showed a smaller, admin-scoped model list, to avoid citing a number that doesn't match what a reader's own account might show).

Test plan

  • Frontmatter validates against the content collection schema (slug, lang, title, summary, category, date, cta, cover, draft)
  • Cover image rendered from SVG source at 1200x630, headline names the actual capability (automatic model delegation) directly
  • All four language files created with the same slug; code blocks (hermes config commands, delegation task prompt) and every quoted UI/tool-output string byte-identical across languages
  • Adversarial review agent confirmed pricing, Hermes's own public docs (delegate_task, delegation.model/provider), and the Cursor no-op citation against source
  • All video/screenshot assets verified to exist; screenshot checked for PII (clean, uses Token Station's Usage page which doesn't show the signed-in account email)
  • No stale $1 credit or unverified catalog-size claims
  • sitemap.xml has all four language URLs

🤖 Generated with Claude Code

Walks through routing Hermes Agent (Nous Research) through Token
Station via a custom OpenAI-compatible endpoint, using OpenAI's
GPT-5.6 family. Centerpiece: real model delegation confirmed working,
Sol plans and delegates a bounded subtask to Luna via delegate_task,
and both models show up billed separately on Token Station's dashboard
in the same session, unlike Cursor's subagent model: field, which our
earlier article documented as a no-op.

Also confirms real tool-use (a file actually written and executed, not
just described), which was the same trap that hit the Cursor+OpenAI
article before its fix landed.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@juntao
juntao merged commit 0cda3e0 into second-state:main Sep 30, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants