agentos.guide › blog
Hermes Agent prompt caching was rebuilt around one lifecycle controller in v0.23.3 on 20 November 2026, and this guide gives you the settings plus three Agent OS prompts to check it is saving you money.
The controller warms the cache through long tool runs, compacts idle chats before the one-hour expiry and makes auto the default, per the release notes.
Background: release v0.23.2 (16 Nov) added the per-session clock this controller reads.
prompt_caching:
cache_ttl: "auto" # default; or "5m" / "1h"; "off" disables
route_ttl_ceilings: {} # e.g. {nous: "5m"}
Auto starts typed chats on 5 minutes and moves to 1 hour after your first pause over five minutes; subagents, cron and API sessions stay on 5 minutes (Hermes docs).
Anthropic bills a long-tier cache write at double the normal input rate and a short-tier write at a quarter more (Anthropic docs), which is why the long tier only pays when your turns are spaced out.
the cache/pruned folder inside your Hermes home.Rank in AI Search or It's FREE →
🔥 Run Hermes inside your own Agent OS
Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members.
Run /context in each of my open Hermes sessions. For each, report
the prompt cache line (warm until, tier, proven/declared/assumed)
and any "Keep-alives: off" note. Flag sessions that are unknown.
Open my Hermes config.yaml. Report the prompt_caching block.
If cache_ttl is pinned to "5m" or "1h", explain the trade-off
against "auto" for my usage and suggest the change. Do not edit
the file until I say yes.
From my provider usage this week, list cache write tokens, cache
read tokens and uncached input tokens per day. Compare with last
week and say whether caching is paying off. Save as a dated note.
Qwen Cloud caps at 5 minutes, OpenAI-compatible routes are treated as 30 minutes, Responses API sessions are never warmed and local models never expire on the clock.
Choosing a model? See our best Hermes Agent model guide.
Join the AI Profit Boardroom →
The cache health check, because it shows every session's tier and warm-until time.
Not unless you approve; the prompt asks it to wait for your yes.
Cache reads rising relative to uncached input on long sessions.
No, local endpoints never expire on the clock.
Run Prompt 1 now and Prompt 3 every Friday.
For more agent routines, join the AI Profit Boardroom, and keep this page as your Hermes Agent prompt caching playbook.
About Julian
I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.
→ The Best Hermes Agent Model, And The Five Model Lanes I Route Inside My Agent OS
→ Hermes Agent Subagents: The Quick Start, And How They Fit Your Agent OS
→ The Best Hermes Agent Setup, And How It Becomes Your Agent OS
→ The Best Hermes Agent Plugins: My Top 10, And Where Each One Fits In Your Agent OS
The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.
Join the AI Profit Boardroom →📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉