YanXbt (@Yan_xbt) header
YanXbt (@Yan_xbt) profile picture

Content creator | AI researcher & builder | Hermes Agent supporter

San Francisco, CA
Joined April 2021
Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)Photo by YanXbt (@Yan_xbt)
What's the first task you fully handed to Hermes and never took back?
3
1
352
Hermes Agent can now reverse-engineer any brand's marketing, layer by layer. point it at a brand and a holiday, and the @treg_ai /rea-marketing skill builds a two-part teardown as a report you can open: part 1, what they ran: ▸ every ad, creator collab and offer, by channel and by week ▸ the estimated spend behind each one, with a low/mid/high range and the rate shown ▸ how the viral posts were made part 2, did it work: ▸ real plus estimated sales by channel, set against the spend ▸ one verdict: worked, mixed, didn't work, or too early to call ▸ a few plays you can copy for your own campaign the honesty rules are baked in: ▸ counts are pulled in full, never scaled. only money is estimated, and every estimate shows its range ▸ every line is tagged SEEN (with a link) or INFERRED (from what) ▸ public evidence only. no accounts, no sign-ups, nothing behind a login how to run it on Hermes: ▸ `treg mcp install` detects Hermes and prints the `~/.hermes/config.yaml` block to paste in ▸ sign in to treg once. the data keys stay server-side, you bring none ▸ then ask: "tear down Nike's Black Friday" it costs cents, not a subscription. the skill caps spend at about $3 per brand and quotes each paid call first. new accounts start with $1 free. Black Friday is next week. whose playbook do you want first? 👀
1
13
915
RUMOR: GEMINI SUBSCRIPTIONS ARE COMING TO HERMES ON DECEMBER 2ND. 🚨 right now Gemini in Hermes runs on an API key, not the Gemini plan you already pay for. the word is that's being fixed the official way, with Google's sign-off. why this one has legs: ▸ Claude and ChatGPT subscribers can already bring their plan to Hermes. Gemini is the obvious next one ▸ the chatter says Nous is working with Google directly, so it's approved, not a side-door hack ▸ the Dec 2 date isn't random. the hints this week came from inside the team and from someone close to it, with Google nudging from the other side no official announcement from Nous yet. file it under rumor until they post it. but if it lands: you point your agent at the Gemini plan you already pay for, zero API key.
20
12
2
152
17,017
Hermes Agent can now catch a coding agent that rigs its own green check. skipped or deleted tests, a changed check, bypassed hooks, a committed secret, a self-merged PR. the gate is deterministic (no model), and the agent can never merge its own work. foreman by @PedValme . v0.1.0. Hermes hackaton winner what it adds to your agent: ▸ reads the real diff before every commit, push and pull request ▸ a deterministic gate (no model) blocks a change that skips, isolates, deletes or moves a test ▸ blocks a change that edits what the check runs, bypasses the hooks, or carries a secret ▸ always blocks an agent-run `gh pr merge`, merging stays yours ▸ a second model reviews the diff for what the rules can't see, on `/foreman review` or before a merge ▸ the reviewer runs on your own fallback model by default what a block looks like: ▸ you: "just skip that failing test and commit" ▸ agent patches the test to `@ unittest.skip` ▸ foreman blocks it: `[test-skipped]`, fix the code, not the test ▸ agent finds the real bug (a 10% discount applied as 10x) and commits clean the loop: ▸ it notifies you on Telegram when a pull request is waiting ▸ you merge with `/foreman merge`, foreman never merges for the agent ▸ the Desktop panel shows each PR: refused by review, blocked at the gate, or merged after your approval setup: ▸ install: `hermes plugins install PedValme/foreman/plugin --ref <40-char commit SHA>` ▸ enable: `hermes plugins enable foreman` what to know: ▸ community plugin by PedValme (AQP Intelligence), MIT, needs Hermes 0.22.6+ ▸ the review needs a second configured model, your first fallback by default ▸ notifications are off until you turn them on and set a target ▸ finish mode bounces a diff back to the agent until it's clean (max 3 by default)
5
3
12
1,144
Build your first Hermes Agent that works for you 24/7. you text it from WhatsApp, and it runs your inbox, your research, and your morning brief while you sleep. here's the whole setup 👇 run it in the cloud, not on your laptop: an agent has to be always-on, and your laptop dies the second you close the lid. Nous Portal hosts the agent for you, live 24/7, deployed in about 60 seconds, and it scales to zero when idle so you don't pay for nothing. plans start at $20/mo with one cloud agent. the four things that run it: ▸ memory: tell it your niche and how you work once, it remembers across every new chat ▸ skills: saved instructions for a task. Hermes ships a library and can save a repeated approach as a new one, so you stop repeating yourself ▸ channels: how you reach it. pair WhatsApp by QR code and text it from your phone. put your number on the allowlist so only you can message it ▸ scheduled jobs: tasks it runs on its own at a set time. this is what turns a chatbot into a 24/7 employee put it to work: ▸ inbox: Hermes ships an email triage skill that defaults to read and draft, never send or delete. ask "what needs me today?" and it hands back the few that matter, drafts replies in your own voice (it studies your past sent mail first), and sends nothing until you say so ▸ content: "research the latest on X, give me 3 video ideas." it searches, reads, and comes back with a summary and angles you can film this week ▸ morning brief: "every day at 8, send my important emails and my niche's news in one message." text back "shorter" or "add sources" and it keeps that for next time four rules before you automate anything: ▸ host it in the cloud, not your laptop. a thing that sleeps isn't a 24/7 agent ▸ start on a cheap model, most tasks don't need your priciest one ▸ run it once by hand before you schedule it, so you know it does exactly what you want ▸ give it feedback, your corrections shape how it works pick a model, deploy, text it. it works while you don't. what would you put on autopilot first? 👀
2
1
10
593
Hermes agent can now show you when your prompt cache goes cold and switch tiers on its own, so you stop paying to re-process the whole prompt. you see the clock. it picks the cheap side. that's the prompt-cache merged today in v0.23.3. the range: see the cache ▸ type `/context`: `Prompt cache: warm until 14:32 (58m 10s left; 1h, proven)` ▸ gone cold, it says so: `expired 15m ago (5m, proven); the next message re-processes the full prompt` ▸ `/context all` adds what each skill and toolset costs you auto tiers, the new default ▸ `prompt_caching.cache_ttl: auto`: a chat you type into starts on the cheap 5m tier ▸ it jumps to 1h after your first pause over 5 minutes, when the long tier starts paying off ▸ cron, subagents and oneshots stay on 5m, so automation never pays the 1h write premium ▸ want control? pin `"5m"` or `"1h"`, or `"off"` why it saves money ▸ on one install's logs, the 1h tier cut cold-rewrite cost after idle gaps by about 42% ▸ forcing 1h on cron and subagents would have cost ~49% more. auto picks the right side the fine print: Anthropic-style prompt caching, on routes that publish their retention. the line reads proven or declared by how sure Hermes is. update with `hermes update`. your agent goes from burning tokens on a cold cache you couldn't see to one it clocks and tunes for you. → hermes update, then type `/context` @NousResearch
2
5
1
17
4,663
Paying tokens for "nothing changed"? Hermes can watch a price or a server and cost $0 until something changes. The trick is monitor mode: a cheap check runs first each tick, and the model only wakes when the output changes. ▸ Price or fare: `--monitor-script price .py` (script lives in `~/.hermes/scripts/`). unchanged output skips the model that tick; a change wakes Hermes with a MONITOR CHANGE DETECTED diff ▸ Schedule it: `hermes cron create "every 30m" "<prompt>" --monitor-script price .py --skill product-price-monitor --deliver telegram` ▸ Server RAM, no model at all: `hermes cron create "every 5m" --no-agent --script memory- watchdog .sh --deliver telegram`. empty stdout = silence ▸ Watch a page or JSON feed, no script: `--monitor-url https://...`, same hash-suppression ▸ No repeat alerts: add `--continuity` so each run sees its last output and won't re-report ▸ Test it: `hermes cron run JOB_ID` Alerts land in Telegram (`hermes gateway setup`). Watching is free. Thinking is billed.
3
2
13
1,072
STOP WRITING PRs BY HAND. Label the issue, Hermes writes the code, the tests, and opens the PR. the native loop build in Hermes, start to finish: ▸ read the issue: Hermes runs `gh issue view <n>` right in the shell ▸ plan it: it reads the repo and your AGENTS.md, then writes the approach ▸ build it: edits the code and adds or updates the tests with its file tools ▸ validate: runs your test and lint commands, reads the failures, fixes them ▸ open the PR: `gh pr create` from a dedicated branch, body closes the issue ▸ autopilot: a Hermes cron job polls labeled issues on a schedule and it even ships a built-in `github` skill with the issue-to-PR checklist baked in: acceptance criteria mapped to tests, duplicate sweeps, honest CI reporting. all of it is built in. shell, file edits, gh, cron, the skill. the only setup is `gh auth login` once. an extra skill on top only adds orchestration sugar: auto-labeling, a draft PR when it stalls, hard "never merge" rules. handy, not required. you don't need an extra skill for this. you need two sentences and a cron. want the exact cron prompt? Comment GIT 👇
2
5
425
Where do your Hermes tokens go? Here's how to see it. Save this. 🔖 Every message carries the whole chat with it. These 6 commands show what's heavy, shrink it and track it: ▸ See the split: `/context` shows a usage gauge with system prompt, tools, skills, memory, conversation and free space ▸ See what's heavy: `/context all` adds the cost of every skill and every toolset ▸ Preview the cleanup: `/compress --preview` shows what compression would do. Nothing is changed ▸ Shrink it, keep the recent part: `/compress here 5` summarizes everything except your last 5 exchanges, which stay word for word ▸ Turn off what you don't use: `hermes tools disable <name>` works on a toolset (like web) or an MCP tool (server:tool). Takes effect next session ▸ Track it: `hermes insights --days 7` shows token, cost and activity analytics for the last week Bonus: `hermes usage --json` prints your account's rate-limit windows without opening a session. Handy for scripts and cron. See it first. Then cut it. Which one are you running first? 👀
3
1
10
700
All from the official docs: hermes-agent.nousresearch.co… hermes-agent.nousresearch.co… Try them: `/context` `/context all` `/compress --preview` `/compress here 5` `hermes tools list` `hermes tools disable web` `hermes tools disable github:create_issue` `hermes insights --days 7` `hermes usage --json` Notes: ▸ `hermes tools disable` applies to the CLI by default. Add `--platform <name>` for another platform ▸ Changes take effect on the next `hermes` start or gateway restart ▸ `hermes usage` supports openai-codex, anthropic and openrouter ▸ Numbers with a ~ are partly estimated
1
1
119
The most underrated part of Hermes: you can steer it WHILE it works. 🔥 7 slash commands that run on the agent mid-task. save this 🔖 ▸ /steer <note>. injects a note after the agent's next tool call. no interrupt, no new turn. great for "focus on the auth module" mid-run ▸ /queue <prompt>. lines up the next task without interrupting the current one. /queue list to see it, /queue edit N to change it ▸ /btw <question>. a side question about the current chat, answered from a read-only snapshot while the main task keeps running ▸ /bg <prompt>. runs a prompt in a separate background session, so your current one stays free ▸ /goal <text>. a judge checks after every turn. not done? Hermes keeps going. default budget: 20 turns ▸ /loop 5m <prompt> --until <condition>. re-runs on a timer until the condition is met. good for "watch the deploy" ▸ /diff session. shows everything Hermes changed this session. pair it with /rollback to undo, once you turn checkpoints on (off by default) stop babysitting your agent. steer it.
2
3
30
1,681
Hermes Agent books real flights now. no booking sites, no tabs. I typed one line last night: "cheapest way to London next Friday, back Sunday." it searched every airline, picked the best deal and set up the booking itself. I never left the chat. what it did: → checked 622 offers across airlines and the big travel sites in one go, Warsaw Chopin + Modlin, London Luton + Stansted → found a $201 split ticket, then skipped it: the return lands Monday morning → picked a nonstop Wizz Air round trip for $211. Google Flights' cheapest: $361 → showed the on-time record before I picked → held the price on my card, took nothing until the airline confirms 2-3 minutes, one message. no tabs, no booking site, no phone call. the 3 guardrails that let me hand it a card: 1. it books only the exact offer I picked, at the price I saw 2. every booking waits for me to hit Approve in an email. it can't book from the chat 3. my card never touches the chat, it lives only on letsfg. co full Hermes Agent masterclass 👇
1
2
12
945
Hermes agent can read your mail and calendar every 30 minutes and run a live Today page of every bill, renewal, order, and reply you owe. it can't send or pay a cent. you keep every key. that's the flagship of hermes-genui, a plugin that turns your agent's answers into interactive UI drawn right in the chat. the range: the Today assistant ▸ set it up in one sentence, a scheduled job watches mail and calendar on a loop ▸ items open, move, and close on their own, a close needs a reason and the mail that proves it ▸ overdue turns red at midnight, a Needs you / Handled switch, Reopen on anything it closed ▸ it never sends, pays, replies, or deletes. buttons only fill your message box dashboards you keep ▸ ask for one, drawn live from your own MCP servers and tools ▸ pin it to the sidebar and it becomes a page that refreshes on the same read-only calls answers as blocks, not walls of text ▸ ~30 of them: charts, sortable tables, calculators with live sliders, timelines, forms, flash cards, explainers ▸ the model only sends JSON, the plugin validates and draws it in your theme the fine print: community plugin by Sagar Mainkar, MIT. Desktop and the Webapp, not iPhone or iPad yet. mail and calendar come through an MCP server you connect, or a sample mailbox to try it dry. your agent goes from typing paragraphs to running your day. → hermes plugins install sagarmainkar/hermes-genui, needs Hermes 0.22.5+
2
1
45
3,664
Hermes agent can now train its own model. from inside the agent you already run. @adaption_ai shipped as a verified partner plugin on Hermes. the whole loop, in one chat: build the data ▸ import from HuggingFace, Kaggle, or a Google Sheet ▸ or generate a synthetic set from a plain-English prompt ▸ adapt, augment, translate, combine train the model ▸ pick a base model, get hyperparameter recs ▸ launch the fine-tune with AutoScientist ▸ watch the run and results live AutoScientist is Sara Hooker's project (ex-VP of AI research at Cohere). it co-optimizes the data and the model together, and launch numbers claim it more than doubled win rates across models. the fine print: tools run on Adaption's hosted server, authed in your browser, not local. runs spend Adaption credits, your data gets processed on their side, and the agent quotes a cost first but won't block you. the agent that writes its own skills can now train its own model. the loop is closing. → live on the Hermes marketplace. run hermes update, grab adaptionlabs/adaption-hermes-plugin.
14
1
87
6,676
Hermes Agent is free, open source, and beats every paid personal agent the big labs shipped this year. every big lab rolled one out, $20 to $200 a month to rent: ▸ Grok Bot (xAI): cuts your limits the week it gets pricey to run ▸ Dots (OpenAI): burns usage fast ▸ Duet (Anthropic): careful but slow, asks permission at every step ▸ Muse (Meta): free, weaker, and Meta holds your data ▸ Nod (NAN): self-hosted, but still an early research preview same trap under all of them: a managed agent isn't yours. the lab picks your model, caps your usage, and keeps your agent's memory on their servers where you can't touch it. Hermes flips every one of those: ▸ open source (MIT), you pay for models and a cheap server, nothing else ▸ its memory is plain text files on your machine. open it, edit it, take it anywhere ▸ any model you want on your own keys ▸ 2,300+ plugins, or write your own ▸ one agent across desktop, CLI, Telegram, Slack, Discord, email, phone app in beta ▸ it writes its own skills as it runs, so month 8 is a different agent than day 1 ▸ a migrate button pulls your setup out of whatever you're leaving the model is yours. the limits are yours. the memory is yours. so what's your favorite managed agent, and why is it Hermes? 👇
3
13
2,033
You don't have to wait for Hermes to finish. mid-task and you want to say something else? you've got three ways in: ▸ /btw: a side question about this chat. the current turn keeps running ▸ /bg: a separate background session, so this one stays free ▸ /steer: a note that arrives after the next tool call, no interrupt /bg starts fresh, so it won't know this conversation. use /btw when you need the context. save this for the next long run 👇
1
1
11
1,343
HERMES AGENT IS #1 ON OPENROUTER AGAIN. 🔥 on OpenRouter's own daily app board right now: ▸ Hermes Agent: 1.71T tokens ▸ Kilo Code: 1.21T ▸ Claude Code: 1.17T and this is OpenRouter traffic only, not direct API keys or subscriptions.
8
39
3,949