ClawStation (@getclawstation) profile picture

Managed hosting for OpenClaw and Hermes. A real machine, not a sandbox: your own network-isolated container with a root shell. Built by @AjeyGore.

World
Joined March 2026
Photo by ClawStation (@getclawstation)Photo by ClawStation (@getclawstation)Photo by ClawStation (@getclawstation)Photo by ClawStation (@getclawstation)
ClawStation is a plain box that runs OpenClaw or Hermes. Dedicated container. Shell access. Your own API keys. Flat price. That's the whole pitch. Thread on why that's the pitch ↓
1
1
2
640
OpenClaw’s web UI now works offline and uses a smart websocket compression/reconnect logic. Been flying to Vienna today and got plenty work done, despite spotty airplane wifi.
125
29
16
748
87,016
ClawStation retweeted
Building Together with AI Agents | ClawCast Community Roundtable teapawt.org/i/broadcasts/1qKVmydgy…
2
3
1
67
18,789
your Hermes subagents can only load the same image 3 times per session... but your main agent has no limit at all every time it loads a screenshot natively that image gets re-sent on every later call in the chat, so the same picture looked at again and again keeps adding up one setting gives your main agent the same cap... after that Hermes tells the model to answer from what it already sees and if dense table screenshots come out unreadable there's a second setting that lets images stay bigger no plugin needed, it's all built in check them out below!
4
2
17
1,672
𝗛𝗲𝗿𝗺𝗲𝘀 Agent can get loose OMG ? 𝗛𝗲𝗿𝗺𝗲𝘀 Agent Desktop plugin runs with the app's full power. -It isn’t sandboxed from the app. -It runs with the app’s authority, so treat installing it as trusting code.🤯 Picking from hundreds of plugins? You can tell a safe one in 30 seconds. 𝗛𝗲𝗿𝗺𝗲𝘀 5 checks before you install a plugin 📣 ▸ Who's behind it? Official = Nous, Verified = the real company, Community = anyone ▸ What can it touch? The install prompt lists its tools, hooks and keys first ▸ Where does your data go? Read the disclosure. Some send searches out or spend credits ▸ Twist: a Desktop page runs inside the app with its full power, so trust it like code ▸ Know the off switch: hermes plugins disable <name> · remove <name> Start with: hermes plugins search <name> Friday, Ep 3: 5 features you asked for that already shipped. Which of these do you check today?
1
3
15
1,529
ClawStation retweeted
which settings are worth changing when a local model runs Hermes Agent, and which change nothing you can measure? Qwen3.8-27B on one 32 GB RTX 5090, one setting per run, the same 100 multi-turn tool-call tasks every time. the clip runs one Hermes Agent task twice on the same model, thinking on and then thinking off, shown side by side. thinking off finished in 8.6 s and thinking on in 18.6 s. - 128k context fits: 29.4 GiB of the card's 31.8 with the default f16 cache, 26.1 with q8_0 - q8_0 KV cache: 78 of 100 tasks against 81, no measurable difference - thinking off: 80 against 81, no measurable difference, with 61% fewer tokens - context 32k / 64k / 128k: 81 / 81 / 81 - q4_0 KV cache: 79 against 81, no measurable difference, with decode speed halved at a 120k prompt the same setup run twice reproduced all 100 tasks exactly, so these gaps are the settings and not run-to-run noise. the tasks never fill a long context, so the context rows say what fits and not how accurate the model is at 100k. on a 32 GB card the defaults already hold 128k. use q8_0 when you need the 3.3 GiB, and try thinking off when tool-call turns feel slow. github.com/notwitcheer/llm-b…
10
6
33
2,644
𝗛𝗲𝗿𝗺𝗲𝘀 Agent now installs on Windows like any other app. No WSL, no terminal. 🤯 Still running it inside WSL, or put off by a PowerShell one-liner? This one's for you. 𝗛𝗲𝗿𝗺𝗲𝘀 Desktop is on the Microsoft Store: free, published by @NousResearch, rated 4.7★. The Store keeps it updated for you. How to get it 👇 ▸ Open the Microsoft Store and search "𝗛𝗲𝗿𝗺𝗲𝘀 𝗔𝗴𝗲𝗻𝘁" ▸ Click Get, then open it from Start ▸ Follow the first-run setup to connect a model ▸ Prefer the terminal? Run this in PowerShell: iex (irm hermes-agent.nousresearch.co…) Start with: "What can you do on this PC?" Are you running 𝗛𝗲𝗿𝗺𝗲𝘀 on Windows yet, or still in WSL? @Teknium
3
10
1
79
5,765
Memory is the one Hermes setup everyone argues about. So here are 24 ways to give Hermes a memory, all in the official plugin catalog. Cloud, fully offline, plain notes or shared across every AI you use. Pick yours. The big names ▸ Hindsight github.com/vectorize-io/hind… ▸ Honcho github.com/plastic-labs/honc… ▸ mem0 github.com/mem0ai/mem0 ▸ OpenViking github.com/volcengine/OpenVi… ▸ Supermemory github.com/supermemoryai/her… ▸ Cognee github.com/topoteretes/cogne… Local and private ▸ Mnemosyne github.com/mnemosyne-oss/mne… ▸ Holographic github.com/itpartypattaya/he… ▸ Compartment github.com/MaxFreedomPollard… ▸ SuperLocalMemory github.com/qualixar/superloc… ▸ ReMe github.com/agentscope-ai/ReM… ▸ HyAtlas github.com/tuancookiez-hub/H… Notes, logs and wikis ▸ hwiki github.com/ottosulin/hwiki ▸ Memory Wiki (official) github.com/NousResearch/herm… ▸ GBrain github.com/kerrz2020/hermes-… ▸ OptMem github.com/rarf/optmem-herme… ▸ LCM-X github.com/electricsheephq/l… ▸ hermes-dreaming github.com/itpartypattaya/he… Cloud, shared and coding ▸ Memanto github.com/moorcheh-ai/meman… ▸ BetterDB github.com/BetterDB-inc/moni… ▸ Memoria github.com/matrixorigin/memo… ▸ Membase github.com/unibaseio/membase ▸ deja-memory github.com/vshulcz/deja-vu ▸ Sugar github.com/roboticforce/suga… Save this for the next time someone asks which memory to use.
57
67
13
694
44,658
self-hosting trap: backups that live on the same box and have never been restored. you find out the day the disk dies. on ClawStation every station is backed up daily, and you can take a backup on demand, restore it yourself and download a copy.
1
1
137
ClawStation retweeted
I finally had a need for Xcode and getting it setup on my Mac Mini for remote vibe coding, with fully automated builds shipped to TestFlight via Hermes over Discord. I feel so l33t like @drbarnard now. Now to build some iPhone Duo apps!
6
1
6
511
It's live. The home base for everything Hermes Agent. HermesWatcher.com is a free field guide to Hermes Agent by Nous Research. Plain English, every source attached. Not sure where to start? Answer 5 questions: ▸ Where will Hermes run? ▸ How do you want to talk to it? ▸ What should it help with first? ▸ What matters most for cost? ▸ How comfortable are you with the terminal? You get a setup plan in order, with the right guides and a workflow to match. What's inside: ▸ 40+ guides and fixes ▸ Every plugin in the official catalog, searchable and updated daily ▸ All 211 official skills with copy-paste installs ▸ Step-by-step workflows with worked examples ▸ 130+ answers to the questions people ask most ▸ Every Hermes release explained ▸ A cheat sheet and glossary Built for your agent too. Point Hermes at the site and it can read the whole guide, then answer your setup questions with the page it came from. The Weekly Watch newsletter is coming. Every Hermes change that matters, explained in five minutes. Bookmark it. Send your agent. hermeswatcher.com
22
33
8
425
33,992
ClawStation retweeted
Anthropic's new agent guide: stop overbuilding agent swarms Start with one agent loop. One model, one context, skills and tools loaded on demand. Keep a subagent only when it earns it: - context pollution - real parallel work - specialized tools In one comparison, the single loop ran about 2x faster. This is why I like how Hermes does delegation One orchestrator that plans, and subagents only when a route makes sense: a cheap model for lookups, a deep one for the hard stuff. Fewer handoffs, less telephone. claude.com/resources/article…
5
1
19
2,006
Ctrl+C no longer kills your /review. 🛑 @NousResearch's 𝗛𝗲𝗿𝗺𝗲𝘀 now ends only the turn you stop. In the PR's live test, 1 background subagent was still running after the stop. Before: 0. ▸ Ctrl+C, Esc, Stop button and ACP cancel ▸ Also when a new message interrupts ▸ /stop still ends background work ▸ Closing or resetting the session does too ▸ Desktop Stop keeps running subagents and processes alive Stop the chat, keep the work. 🔥
3
2
10
832
ClawStation retweeted
Generate SFX with @ElevenLabs straight from Hermes Desktop with looped playback, one-click variation generation, and a helpful library of all your sounds.
Introducing the @ElevenLabs SFX plugin, built to continue the work on a suite of Game Dev Studio capabilities for your Hermes Agent! Hermes made this demo video to see it in action, you can install it right now from the plugins catalog: hermes-agent.nousresearch.co… The plugin adds a tool to hook up your ElevenLabs account API and generated sounds for whatever you need. Additionally, on the desktop app, the plugin adds a gallery of all of your generated SFX, easy integration into your game or other multimedia, looping, cost, cycling through the generations for that sound effect, and more! Enjoy
23
12
4
247
30,626
ClawStation retweeted
The OpenClaw v2026.10.8 patch is out 🦞 🔧 Fixes for installs and updates 💬 Fixes for stuck chats and missing replies 📱 More fixes across apps, channels and providers Thanks to all 163 contributors! Full notes docs.openclaw.ai/releases/20…
25
31
9
351
37,540
self-hosting trap: your agent shares one box with a few side projects, a runaway task eats all the memory, and everything on it falls over. on ClawStation CPU, memory and disk are capped per station at the size of your plan.
1
1
145
ClawStation retweeted
Hermes Agent has a skill that finds where data brokers expose your personal info and files the removal requests for you. it's called UNBROKER by @SHL0MS open source. runs locally. your data never leaves your machine. the problem: hundreds of brokers publish your name, addresses, phone, email, birthday, even your relatives. anyone can find where you live in ten seconds. you have the legal right to delete it. CCPA, CPRA, and state privacy laws. but every broker has a different process and many make it intentionally difficult. companies like DeleteMe charge $330/yr to file removals you can submit yourself for free. and you're handing a new company the exact data you want erased. what UNBROKER does: ▸ builds search vectors from everything: every name, alias, email, phone, address you've had. brokers index you under maiden names and houses you left in 2014. ▸ clusters parent brokers first. many sites are resold shells of a few parents. one removal can clear a dozen child sites. ▸ verifies a real listing exists before submitting anything. confirms it's you and not a namesake. sends only the exact fields a broker requires. ▸ automation is tiered (T0 to T3): fully automated → automated with verification → human-verified → human-only. the agent takes the highest viable tier. human tasks accumulate into one digest at the end. ▸ email doesn't need a stored password. sends through your logged-in webmail. or SMTP. or writes drafts for you to send. ▸ California residents: the DROP portal deletes your data from ~545 registered brokers in a single verified request. ▸ encrypted at rest with age. opaque IDs everywhere. your real name never appears in filenames or logs. ▸ brokers relist you eventually. every case tracked in a ledger. set a cron schedule. re-listings get re-filed. ▸ consent-gated. the engine refuses to scan or act without recorded authorization. this is a removal tool, not a people-search aggregator. install: hermes skills install official/security/unbroker then in any session: "use the unbroker skill to remove my data from data brokers. run it hands-off and show me the human-task digest at the end." scope is US-first. EU/UK GDPR coverage is on the roadmap. the engine and autonomous loop are built (v1.0, 85 tests). live submission against broker sites is actively field-tested. $330/yr for DeleteMe. $0 for UNBROKER. open source. local models work.
4
4
18
1,571
10 days of contributing to @openclaw (Oct 16–26): • 10 PRs opened • 10 PRs merged • 5 bug reports filed • 4 issues closed as fixed • 12 more PRs open and in review 🦞The lobster way:🦞 Every fix comes from running OpenClaw for real, every day: find it, reproduce it, fix it, prove it against a live Gateway, send it upstream. The best part of open source: you don't have to live with a bug. You can fix it. What landed: – Claude CLI chats keep their conversation when you stop the very first turn – Session titles show the actual title instead of the model's explanation – Node exec output no longer comes back as a duplicate wake – A round of Control UI polish: sidebar child sessions follow your sort order, setting an icon or color no longer bumps a session to the top, a "no emoji" choice in the icon picker, emoji board icons in the sidebar, cleaner status filters – Workboard: the unfiltered page is now titled "All boards" – macOS computer-use live proof runs through instead of aborting before its first check
4
1
1
19
880