📱 𝗛𝗲𝗿𝗺𝗲𝘀 Mobile beta filled up fast at NousCon. The waitlist is open now. ▸ Nous says they're freeing up more beta seats ▸ They're also getting the app ready for a wider release ▸ Join here: Check: the page loads the waitlist form Limit: no release date announced yet

𝗛𝗲𝗿𝗺𝗲𝘀 Agent was told to fix itself. It pruned 237 sessions and broke its own install. 😬 The owner spent about 2 days repairing it by hand. If 𝗛𝗲𝗿𝗺𝗲𝘀 acts up, don't make 𝗛𝗲𝗿𝗺𝗲𝘀 the mechanic. Run these 5 in order: ▸ hermes backup ▸ hermes doctor ▸ hermes dump ▸ hermes debug share ▸ hermes sessions prune --dry-run Two things it should never do: run git checkout in its own install, or edit state.db by hand. Save this for the bad day. What each command does is in the first reply. Which of the 5 have you never run?

𝗛𝗲𝗿𝗺𝗲𝘀 Agent was told to fix itself. It pruned 237 sessions and broke its own install. 😬 The owner spent about 2 days repairing it by hand. If 𝗛𝗲𝗿𝗺𝗲𝘀 acts up, don't make 𝗛𝗲𝗿𝗺𝗲𝘀 the mechanic. Run these 5 in order: ▸ hermes backup ▸ hermes doctor ▸ hermes dump ▸ hermes debug share ▸ hermes sessions prune --dry-run Two things it should never do: run git checkout in its own install, or edit state.db by hand. Save this for the bad day. What each command does is in the first reply. Which of the 5 have you never run?

28 paid model calls a week just to say "no change." 𝗛𝗲𝗿𝗺𝗲𝘀 Agent skips them. 💸 Check a page every 6 hours and a normal watcher calls the model 28 times a week just to say nothing happened. 𝗛𝗲𝗿𝗺𝗲𝘀 Agent checks the page first (hermes cron create … --monitor-url). Same page? The run stops. No model, no cost. Page changed? Only then does the model wake, and it gets just the difference. 28 paid "no change" answers vs 1 useful call. It's setup 01 of the 10 in my 𝗛𝗲𝗿𝗺𝗲𝘀 article (v0.22) 👇 What page would you have 𝗛𝗲𝗿𝗺𝗲𝘀 watch?

28 paid model calls a week just to say "no change." 𝗛𝗲𝗿𝗺𝗲𝘀 Agent skips them. 💸 Check a page every 6 hours and a normal watcher calls the model 28 times a week just to say nothing happened. 𝗛𝗲𝗿𝗺𝗲𝘀 Agent checks the page first (hermes cron create … --monitor-url). Same page? The run stops. No model, no cost. Page changed? Only then does the model wake, and it gets just the difference. 28 paid "no change" answers vs 1 useful call. It's setup 01 of the 10 in my 𝗛𝗲𝗿𝗺𝗲𝘀 article (v0.22) 👇 What page would you have 𝗛𝗲𝗿𝗺𝗲𝘀 watch?

𝗛𝗲𝗿𝗺𝗲𝘀 v0.22.7 just dropped. The 1 feature worth knowing: a lead bot runs the room. 🔥 Before: one message that names nobody, and every bot in the room wakes up. Now: the lead reads it and hands it to the right one. ▸ Room settings → "Who answers a message that names nobody" → Lead ▸ Test run: "ada gave this to linus" in 3.37 s. scout never woke. ▸ Cron jobs post straight into the room and wake only who they @mention: hermes cron create 60m --name morning-check --no-agent --script --deliver room:@default Check: cron post landed 0.77 s after hermes cron run, 1 of 3 bots woke Limit: Lead is opt-in, the default is still "Everyone answers" Run hermes update to get it.

𝗛𝗲𝗿𝗺𝗲𝘀 v0.22.7 just dropped. The 1 feature worth knowing: a lead bot runs the room. 🔥 Before: one message that names nobody, and every bot in the room wakes up. Now: the lead reads it and hands it to the right one. ▸ Room settings → "Who answers a message that names nobody" → Lead ▸ Test run: "ada gave this to linus" in 3.37 s. scout never woke. ▸ Cron jobs post straight into the room and wake only who they @mention: hermes cron create 60m --name morning-check --no-agent --script --deliver room:@default Check: cron post landed 0.77 s after hermes cron run, 1 of 3 bots woke Limit: Lead is opt-in, the default is still "Everyone answers" Run hermes update to get it.

Three jobs most people do badly with an AI agent: audit code, design pages, hunt for work. 😬 3 open-source repos (172K stars combined) give 𝗛𝗲𝗿𝗺𝗲𝘀 Agent a playbook for each (v0.22): 1. security-audit (Cloudflare, 57.3k ⭐): a verifier tries to disprove every bug found. Only what survives lands in REPORT.md. 2. hallmark (Together AI, 45.9k ⭐): builds pages that don't look AI-made, then checks 8 screen widths. 3. ai-job-search (69.1k ⭐): scores job posts, tailors your CV, writes cover letters. Limit: they're community skills. Run hermes skills inspect before you install. Which of the 3 is your biggest time sink right now?

65 new 𝗛𝗲𝗿𝗺𝗲𝘀 plugins joined the catalog since Monday. These 5 fix problems you actually have. ✈️ 1. Long terminal output eating your tokens? tokenjuice by @vincent_koc shrinks it before the model reads it. 2. A task stuck waiting on you? kanban-empowered by @dellanio asks you on Telegram. Reply to unblock it. 3. Paying per track for background music? hermes-band by @RightSideOfAI writes and renders it on your machine. No music API. 4. A clip and no idea where it was recorded? aircraft-doppler-geolocation by mojomast narrows it down from a passing plane's sound. 5. Not sure where an answer came from? search-sources by ShowUNow lists every page it searched, in a Sources panel in 𝗛𝗲𝗿𝗺𝗲𝘀 Desktop. Install any: hermes plugins install They're community plugins, so run hermes plugins info first. Tap the video for all 5 in 30 seconds 👇 Which problem do you hit most?

📱 𝗛𝗲𝗿𝗺𝗲𝘀 Mobile beta filled up fast at NousCon. The waitlist is open now. ▸ Nous says they're freeing up more beta seats ▸ They're also getting the app ready for a wider release ▸ Join here: Check: the page loads the waitlist form Limit: no release date announced yet

Thank you to everyone who traveled from across the world to NousCon 2026. We loved spending time with the people who make the Hermes community great. As you all know, we released the Hermes Mobile beta at the event, and as anticipated it filled up quickly. We are working tirelessly to free up more room in the beta, and to get the app production ready for wider release. The feedback has been incredible so far, and we are excited to share it with the world soon. In the meantime, please join the waitlist at

65 new 𝗛𝗲𝗿𝗺𝗲𝘀 plugins joined the catalog since Monday. These 5 fix problems you actually have. ✈️ 1. Long terminal output eating your tokens? tokenjuice by @vincent_koc shrinks it before the model reads it. 2. A task stuck waiting on you? kanban-empowered by @dellanio asks you on Telegram. Reply to unblock it. 3. Paying per track for background music? hermes-band by @RightSideOfAI writes and renders it on your machine. No music API. 4. A clip and no idea where it was recorded? aircraft-doppler-geolocation by mojomast narrows it down from a passing plane's sound. 5. Not sure where an answer came from? search-sources by ShowUNow lists every page it searched, in a Sources panel in 𝗛𝗲𝗿𝗺𝗲𝘀 Desktop. Install any: hermes plugins install They're community plugins, so run hermes plugins info first. Tap the video for all 5 in 30 seconds 👇 Which problem do you hit most?

Three jobs most people do badly with an AI agent: audit code, design pages, hunt for work. 😬 3 open-source repos (172K stars combined) give 𝗛𝗲𝗿𝗺𝗲𝘀 Agent a playbook for each (v0.22): 1. security-audit (Cloudflare, 57.3k ⭐): a verifier tries to disprove every bug found. Only what survives lands in REPORT.md. 2. hallmark (Together AI, 45.9k ⭐): builds pages that don't look AI-made, then checks 8 screen widths. 3. ai-job-search (69.1k ⭐): scores job posts, tailors your CV, writes cover letters. Limit: they're community skills. Run hermes skills inspect before you install. Which of the 3 is your biggest time sink right now?

Your 𝗛𝗲𝗿𝗺𝗲𝘀 can turn off the lights, queue your music and clean up your call audio. 🤯 𝗛𝗲𝗿𝗺𝗲𝘀 101 · Ep 1: 5 plugins from the official catalog, one per everyday job: 1. Home Assistant ▸ See every device and its state ▸ "Turn off the living room lights" Needs Home Assistant and its token. 2. Spotify ▸ Play, pause, skip and queue playlists on any device Needs your own Spotify app. Premium for playback. 3. TinyFish: research with sources (needs a key, search is free) 4. Memory Wiki: find what you decided (no setup) 5. NVIDIA Broadcast: better calls (Windows + Broadcast 2.3+) Which job would you hand over first?

Hermes Agent in one sentence, and 3 commands that prove it. 🧪 An AI agent that uses tools, remembers context, and builds reusable skills. Miss one of the three and you're using it like a chatbot. So I checked all three on my own machine: ▸ Uses tools: 26 toolsets on. Terminal, files, browser, search, cron ▸ Remembers: MEMORY.md and USER.md carry into every new session ▸ Builds skills: 30 local skills, saved once and reused Try it on one small chore: "The nightly build failed. Find out why, fix it, and post a summary to #eng#. Save the steps as a skill." Built by @NousResearch and @Teknium. The video shows real output from my machine. The chore prompt is illustrative. The 3 commands to check your own setup are in the first reply. 👇

Stop Losing Hours To Hermes Errors That Already Have A Fix. I read 1,000 Hermes Agent GitHub issues from the last 30 days. Four problems keep coming back. Each has a one-line fix. ▸ Your cron job "ran," but nothing arrived. Update to v0.21.6+, then run: hermes gateway restart ▸ One rate-limit error (429) and your provider is "exhausted" for hours. hermes auth reset ▸ hermes update says "could not resolve the stable source channel." git fetch origin --tags && hermes update ▸ After an update, your memory provider is "not installed." hermes plugins install I checked them against the v0.22.2 docs on Oct 19. They come from the issue threads, not my own tests. 8 more fixes, with issue numbers, are in the first reply.

Stop rewriting your Hermes agent’s job description. “Check the evidence.” “Explain the trade-offs.” “Don’t jump straight to a fix.” If you keep repeating those instructions, a reusable profile is worth exploring. @HermesWatcher surfaced a collection of 68 Star Trek-inspired Hermes profiles. Here are five starting points, matched to actual work: ✦ Spock → investigate a recurring failure. ✦ Data → examine messy data or plan a schema. ✦ Picard → compare options before a difficult decision. ✦ Worf → review permissions and operational risks. ✦ Quark → examine pricing and negotiation trade-offs. Each packages instructions for how Hermes approaches the work, including strengths, blind spots and safeguards. Try one on a task you already understand. Keep the model and input the same, start a fresh session, then compare it with your usual setup. Check three things: what it missed, what you had to correct, and whether the next action was clearer. A profile changes the instructions. Better results still need evidence. Credit to the community contributors and @witcheer for sourcing the collection, hosted on @Teknium’s GitHub. Which job do you keep having to explain to Hermes?