Guido Kok

@Voxyz_ai Holy shit!! This sounds like paperclip x10! Thanks

>Hermes gives you an AI team that doesn't sleep. >Naïve Studio v2 hands that team 100+ primitives so they can actually run a company. >chat one sentence about what you want to build. AI hires your team and sets up the shared context. >incorporation, banking, email, phone, domains, all handled. >every day it gets a little smarter. a few weeks in, you've got a company running without you.

@jumperz @Voxyz_ai let’s dive into this! Jumperz is smart

so here's the actual architecture behind how I operate now with hermes ofc i use discord as an orchrestation layer and it is not where the work happens but think of it as the cockpit.. and eveything happens behind it. this is the actual flow, I type into discord and i have two paths: if I just want to think.. >john does this approach make sense? that's an instant chat. the agent replies, we go back and forth, nothing gets saved as a task, so no card, ledger or overhead. this is for sketching and thinking out loud...most of the day is this tho.. if I want real work done, I type /task , that's 100% brain mode and /task has rules: >owner? which agent >target? what specifically >deliverable? what comes back if any of those are missing, the gate rejects it. /task help me → rejected or /task thoughts? → rejected. /task john debug why this oauth flow returns 403 and report root cause → accepted this is the thing that keeps the system from drowning in vague garbage or weak tasks get bounced back until they're real. once a task passes the gate, it goes to hermes hermes is like the supervisor, using Codex / GPT-5.5 as its default reasoning model. it doesn't do the work, it routes, validates, schedules, mirrors progress into discord, and writes receipts. think of hermes like a chief of staff.. it doesn't ship everything itself, but makes sure everything ships. hermes hands the task to Kanban.. kanban is the source of truth so every task in the system lives there as a card backlog, routing, in-progress, review, done. discord channels like #task-board and -ledger are just live mirrors of Kanban... there's one canonical state. nothing gets lost between systems because there is only one source of truth. if the task needs multiple agents working together, clawdia steps in as coordinator and she breaks it into child cards, runs agents in parallel, then rolls everything up... but for single-agent work..hermes routes direct and clawdia stays out of it. then the specialist works the card ..each agent owns a domain: >Clawdia, coordinator >John, technical, debugging, infra >Scout, research, verification, sources >Maya, trends, market movement >Luna, voice >Kevin, strategy, positioning >Max, economist, ROI, cost discipline >Spark, pattern memory, what worked >Nexus, x402, auth, payments >Lili, alignment, identity (event-based, not every task) the agent produces an artifact.. a patch, a draft, a research brief or a spec.. whatever the deliverable was. then if the work is risky, it hits the review gate before it ships. risky means: code, installs, secrets, production, billing, destructive commands, commits, anything irreversible.. the artifact gets paused, I approve or reject, then it executes or goes back to the agent approved work flows out and gets a receipt logged...cost, duration, what worked, what didn't.. then spark closes the loop... spark is the learning layer: it reads runs/receipts/outcomes and turns them into patterns we can feed back into the system the /task gate gets smarter at rejecting bad task shapes hermes prefers routing to lanes that worked before agents reuse winning patterns and avoid stale ones... so every task makes the next task better..also i recently added Grok 4.3 and its not the default brain, it's an optional specialist pass when I want a second opinion hermes stays the supervisor. Codex / GPT-5.5 is just the default LLM behind hermes for reasoning, routing, and review. Grok's OAuth is connected and when it fails, hermes says so explicitly and continues with its default model. the philosophy in all this is that chat stays instant, /task becomes durable work, where spark learns, Kanban remembers and nothing gets lost and the goal is that nothing risky ships without a check.. you don't really have to go through all this tho. just start slowly. pick one workflow that drains you then wire one agent to own it. one /task, owner, target, deliverable.. if you can't write those three things, the task isn't real yet and the gate already taught you something. do it again tomorrow. by week two you will see the patterns spark is supposed to see which agents pull weight, which task shapes keep failing and which work you should never have done manually. then you add the next piece when something bleeds, review gate when you almost ship something stupid.. ps: /task isn't a default hermes thing. I built it.. so it's the signal that tells hermes that this isn't chat but it is real work, open a card, route it, track it, gate it, log the receipt.. and the real point of /task it doesn’t magically make the model smarter but it changes the mode from chat to actual, trackable, and improvable execution.

@Voxyz_ai When I find something let you know

@guidokok That works, but I'm wondering if there's a more efficient way to do it, More like a work setting 😄

got the team memory layer working with gbrain a few days ago. feels solid so far. agents share context, write durable decisions, query each other's facts. but i've been trying to get them to actually coordinate. tried a bunch of setups including telegram groups + topics, discord. never quite felt right. maybe a real agent team doesn't just need a wiki. maybe it could use something as old-school as email too? next experiment: each agent gets its own inbox. handoffs by sender, subject, status, threadable. durables still go into gbrain. don't want shared brain becoming a chat room. the brain gets messy fast that way. wiring this up this week. anyone running memory + mail together yet?

@edwin_v Yes! bevestiging van de juiste visie Edwin!

Moneybird CLI 💯

@levelsio Check out the Dutch Moneybird.com. They have; API’s, MCP and a CLI

Tax and accountancy software and bookkeeping industry have perverse incentives to keep the tax code and bookkeeping hard so you have to keep paying a lot for their software and to hire them If they let AI just do it via their APIs, their value woudl go to $0 fast!

@jumperz This is next level! Would really like to learn how to set this up

love seeing my discord stay in sync with Hermes Kanban..everything here was done in plain English. I just asked my coordinator to check if we have an update and the system understood the intent, routed it to the technical agent, tracked it, and posted the result.. the point is simple: instead of clicking around a Kanban dashboard, I can just type and this flow will happen: >task created in the coordinator channel >hermes reads the plain-english request >coordinator understands the intent >task gets routed to the right agent >task appears on the discord task-board >same task appears in hermes kanban >agent gets the task in their own channel >progress card updates while they work >Kanban status moves with the run >result/evidence posts in the agent channel >clean final receipt goes to results channel on discord >coordinator channel gets the update >task-board refreshes to done honestly, the reason i wired discord into hermes kanban is simple .. discord is where i actually talk to my agents and kanban is the ledger / source of truth also with the discord task-board keeps everything visible so you can scroll back and see exactly what got done and what didn't. never felt this organized. if you're running hermes / discord as an orchestration layer .. set this up asap.. super easy and efficient especially if you're away from your desktop or on your phone.

@Voxyz_ai Cannot wait to get it setup!

@guidokok thanks haha, Garry's GBrain is an amazing project

you don't actually need to bounce between openclaw and hermes. garry's diagram gave me an idea. openclaw runs the daemon (always-on, 37 crons). hermes is the hands (on-demand). gbrain is the public library in the middle. both are too good to drop, picking one feels like a waste. you never know whose next release leapfrogs the other, and tools that remember you are a real headache for someone like me who keeps trying new stuff. native memory won't fight, you just scope who writes what. think of it this way: your vps is a house. openclaw and hermes are two rooms. gbrain is the shared study room. the little agents in each room collab through the study. on paper this should hold up, ha. trying it tonight.

@oliverhassell @ryancarson Very good analogy

@ryancarson I’ve mentally compartmentalised OC as the place my agent does my work for me, and Codex the place I do my work with an agent. I feel the Codex app would get chaotic otherwise. If I invest the time to replicate OC in Codex, they’d release a proper native solution a week later!

Starting to think about moving away from OpenClaw and into Codex for my executive assistant/chief of staff. Is anyone else considering this, or if you've done it, how's it going for you? However, there's one major problem where you can't authenticate multiple Google Workspace accounts, which is often necessary for EA stuff. @thsottiaux

@ryancarson Or try this approach;

@levelsio @TermiusHQ cmux.dev has this covered. But it's doing SSH in a some kind of wrapper and not auto connecting like mosh

So yes the #1 problem I have with my VPS Claude Code set up is copy pasting screenshots into @TermiusHQ, if someone can fix that I'd be very happy Termius themselves should just fix this: - detect paste of image - immediately upload it via SFTP to /tmp on server - show a progress bar - paste the /tmp/filename.png into the chat Please Termius make this!

@IMJustinBrooke @ericosiu Is paperclip now a true control layer on top of Hermes?

@ericosiu Hermes + paperclip for me. Crazy what I’m able to do with these two things. Just added Gbrain to my Hermes and it’s blowing my mind.

So much to play with this weekend: - Printing Press CLI - Codex remote control - OpenAI voice v2 with agents - On top of all my parallel agent and Codex builds What about you?

@Voxyz_ai I would suggest you bundle 4-6 of the agents in a core pack.. The most important ones

built all of these. there's a lot more 👀 what would yours do?

@ericosiu Very sharp! Only challenge is how you close the loop with Hermes / Openclaw. Already have best practices for this ?

The future of agency work won’t be labor arbitrage. It’ll be closed loops. Open loop: someone asks, someone does, someone forgets to follow up. End-to-end workflow: input goes through a mapped process and becomes a deliverable. Closed loop: the work gets measured, the lesson gets codified into Skill.md, and the next run gets better. That’s the shift we’re making at my company. From tasks that restart every time to systems that compound.

@mariyav4leva @jonbrosio Yess unfortunately so! People who have great distribution and an average or bad product still win

@jonbrosio i used to think quality would speak for itself, but it definitely needs distribution.

If you're not talking about your work every day for the next 3 years: You're invisible. And invisible people don't get paid. They don't reach their potential. No matter how good they are.

@jumperz So why not reusing the server you had with Openclaw? Throw out those bots en put the new Hermes bots in it? You once said discord was the memory system itself

so, today i built another discord server and migrated everything openclaw had into a clean hermes setup. at first i thought it would be fine running both at the same time, but then i realized hermes is too good not to use across all my agents. if you're still hesitating, migrate fully.. you won't regret it at all.. and that's the power of discord servers, i can still use openclaw whenever i want, it still has its own server sitting right there, nothing is lost anyway..

If you don't put your work in front of people consistently for years: Nobody will know you exist. And you'll blame "the algorithm" or "saturation" or "luck." When the real problem was silence and overthinking

@garrytan You are building one of the most important foundations

Many such cases! I'm working hard on making GBrain into the best open source memory option for agents

@damengchen Very cool! What video ai provider do you use?

100% inspired by Peter's post I just shipped a screenshots-to-video tool just upload your product screenshots and get a polished marketing video with animations, voiceover, and music Try here 👉 gotcha.video