# Claude Code ## 19:11:48 - User /youtube-note https://www.youtube.com/watch?v=9G6IPxSXU8s --- ## 19:11:53 - Assistant > [!info]- Thinking > Let me process this YouTube video. I'll start by extracting the video ID and metadata. --- ## 19:11:54 - Assistant ### 🔧 python3 -c ' import re import json import urllib.request import urllib.error url = "https://www.youtube.com/watch?v=9G6IPxSXU8s" patterns = [ r"(?:youtube\.com/watch\?v=|youtu\.be/|youtube\.com/embed/|youtube\.com/v/|youtube\.com/shorts/)([a-zA-Z0-9_-]{11})", r"[?&]v=([a-zA-Z0-9_-]{11})", ] video_id = None for pattern in patterns: match = re.search(pattern, url) if match: video_id = match.group(1) break if not video_id: print("ERROR: Could not extract video ID from URL") exit(1) print(f"VIDEO_ID: {video_id}") oembed_url = f"https://www.youtube.com/oembed?url=https://www.youtube.com/watch?v={video_id}&format=json" try: with urllib.request.urlopen(oembed_url, timeout=10) as response: data = json.loads(response.read().decode()) print(f"TITLE: {data.get(chr(116)+chr(105)+chr(116)+chr(108)+chr(101), \"Unknown Title\")}") print(f"CHANNEL: {data.get(chr(97)+chr(117)+chr(116)+chr(104)+chr(111)+chr(114)+chr(95)+chr(110)+chr(97)+chr(109)+chr(101), \"Unknown Channel\")}") except Exception as e: print(f"TITLE: Unknown (error: {e})") print(f"CHANNEL: Unknown") ' **Status**: failed --- ## 19:12:01 - Assistant ### 🔧 python3 << 'PYEOF' import re, json, urllib.request, urllib.error url = "https://www.youtube.com/watch?v=9G6IPxSXU8s" match = re.search(r'[?&]v=([a-zA-Z0-9_-]{11})', url) video_id = match.group(1) if match else None if not video_id: print("ERROR: Could not extract video ID") exit(1) print(f"VIDEO_ID: {video_id}") oembed_url = f"https://www.youtube.com/oembed?url=https://www.youtube.com/watch?v={video_id}&format=json" try: with urllib.request.urlopen(oembed_url, timeout=10) as response: data = json.loads(response.read().decode()) print(f"TITLE: {data.get('title', 'Unknown Title')}") print(f"CHANNEL: {data.get('author_name', 'Unknown Channel')}") except Exception as e: print(f"TITLE: Unknown (error: {e})") print(f"CHANNEL: Unknown") PYEOF **Status**: completed --- ## 19:12:07 - Assistant ### 🔧 python3 << 'PYEOF' from youtube_transcript_api import YouTubeTranscriptApi video_id = "9G6IPxSXU8s" try: api = YouTubeTranscriptApi() transcript = api.fetch(video_id) formatted = [] for entry in transcript: time = int(entry.start) minutes = time // 60 seconds = time % 60 formatted.append(f"[{minutes}:{seconds:02d}] {entry.text}") print("TRANSCRIPT_START") print("\n".join(formatted)) print("TRANSCRIPT_END") except Exception as e: print(f"TRANSCRIPT_ERROR: {e}") PYEOF **Status**: completed > [!info]- Thinking > Now I have the transcript. Let me generate the content and create the note. Now let me create the note. --- ## 19:12:15 - Assistant ### 🔧 Write Videos/Claude Code Channels Replaced My OpenClaw Setup.md **Locations**: `/Users/nickyoung/My Obsidian Vault/Videos/Claude Code Channels Replaced My OpenClaw Setup.md` **Status**: completed **File**: `/Users/nickyoung/My Obsidian Vault/Videos/Claude Code Channels Replaced My OpenClaw Setup.md` ```diff + --- + tags: + - video + - reference + title: "Claude Code Channels Replaced My OpenClaw Setup" + channel: "Artem Zhutov" + url: https://www.youtube.com/watch?v=9G6IPxSXU8s + date: 2026-03-24 + cover: https://img.youtube.com/vi/9G6IPxSXU8s/maxresdefault.jpg + --- + + # Claude Code Channels Replaced My OpenClaw Setup + +  + + **Channel:** Artem Zhutov + **URL:** https://www.youtube.com/watch?v=9G6IPxSXU8s + **Date processed:** 2026-03-24 + + --- + + ## Summary + + Artem Zhutov demonstrates how Anthropic's new "channels" feature for Claude Code allows him to interact with his Obsidian vault's full skill set and context from his phone via Telegram and Discord. He runs Claude Code from within his Obsidian vault (which he calls "maps world"), loading an MCP server plugin that bridges Claude Code to messaging platforms. This gives mobile access to 150+ custom skills, memory, and CLAUDE.md context that previously required a terminal session. + + The video compares this setup to OpenClaw, an existing multi-platform agent framework. While OpenClaw supports 23+ platforms and has a skill marketplace, Artem prefers Claude Code's approach because of the transparency of local markdown files, the quality of Anthropic's models (especially Opus), and the economics of the $200/month subscription versus API keys. He estimates the token consumption far exceeds the subscription cost. + + He also demonstrates a "loop" pattern for proactive agents (scheduled tasks like Reddit monitoring and therapy-style check-ins), an orchestrator pattern using tmux workspaces for multi-agent coordination, and discusses what's still missing: native multi-agent Discord channels, better observability while agents work, and more platform connectors (WhatsApp, Slack). + + --- + + ## Key Points + + - Claude Code "channels" feature enables phone-based access to your full Obsidian vault context (skills, memory, CLAUDE.md) via Telegram or Discord + - Setup is straightforward: install the MCP plugin, create a bot (via BotFather for Telegram or Discord Developer Portal), configure the token, and run Claude Code with the `--channels` flag + - Compared to OpenClaw: fewer platforms (Telegram/Discord vs 23+), but better model quality, local file transparency, and subscription-based pricing rather than API keys + - The `--dangerously-escape-permissions` flag is used to avoid interactive permission prompts, which is acknowledged as non-ideal + - The "loop" skill enables proactive agent behavior — scheduled tasks at set intervals (e.g., Reddit sentiment analysis every 8 hours, morning briefings every 24 hours) + - Multi-agent orchestration is possible via tmux workspaces: one orchestrator agent can spawn and monitor other agent sessions + - Observability is a current gap — when the agent is working, there's no visibility into its progress until it replies + - The Obsidian vault is the differentiator: skills as playbooks, conversation history, accumulated corrections, and personal context make the agent improve over time + + --- + + ## Detailed Notes + + ### Demo: Claude Code + Discord [0:00–2:43] + + Artem shows a live Claude Code session listening for Discord messages. He asks it to generate a diagram explaining how channels work and to create a PDF of his slides — both are delivered as Discord attachments. The agent has access to his vault's skills (including Excalidraw for diagrams) and can produce and send files directly through the messaging platform. + + ### Why Obsidian Matters [3:17–4:37] + + Obsidian is a markdown viewer over a local folder. Running Claude Code within the vault gives it access to all skills (stored in `.claude/`), memory files, and personal context. Artem's CLAUDE.md describes who he is, guardrails, file organization, writing voice, and more. This rich context is what makes the agent personal rather than generic. + + ### Live Skill Demo: Reddit Analysis [4:37–6:21] + + He asks the agent (via Discord) to use his Reddit skill to analyze sentiment in the Claude Code subreddit, extract posts with upvotes and exact citations, and deliver an inline report with clickable links. After some back-and-forth on formatting, the agent produces a usable report with real thread links and analysis. + + ### OpenClaw vs Claude Code + Channels [6:23–9:04] + + | Feature | OpenClaw | Claude Code + Channels | + |---|---|---| + | Platforms | 23+ (WhatsApp, Signal, etc.) | Telegram, Discord | + | Skills | Marketplace + pre-installed | Local vault skills (full control) | + | Memory | soul.md | CLAUDE.md + Obsidian vault | + | Pricing | API keys (OpenAI or Anthropic) | Subscription ($20–$200/mo) | + | Models | OpenAI default (weaker vibes) | Anthropic (Opus) | + | Observability | Built-in | Limited | + + Artem estimates his $200/month subscription yields thousands of dollars worth of token usage — "heavily subsidized" and unlikely to remain this generous. + + ### Loop Pattern: Proactive Agents [9:06–10:58] + + The built-in "loop" skill allows scheduling recurring tasks. Examples demonstrated: + - Therapy-style check-in every 2 hours + - Reddit/Twitter/YouTube source analysis every 8 hours + - Morning briefing scan every 24 hours at 7am + + These loop jobs deliver results to Discord or Telegram. + + ### Setup Walkthrough: Telegram [11:03–13:07] + + 1. Update Claude Code to version 80+ + 2. Install the Telegram plugin + 3. Create a bot via BotFather, copy the access token + 4. Run `/telegram:configure` and paste the token + 5. Restart Claude Code with the `--channels` flag and Telegram plugin + 6. Set access policy to allowlist-only for security + + ### Setup Walkthrough: Discord [13:07–17:09] + + 1. Install the Discord plugin + 2. Create a new application at Discord Developer Portal + 3. Copy the bot token and configure via `/discord configure` + 4. Set up OAuth2 URL generator with bot permissions (send messages, attach files, read history) + 5. Create a Discord server, invite the bot via the generated URL + 6. Enable the "message content intent" in bot settings + 7. Restart Claude Code with the Discord plugin + + ### Orchestrator Pattern: Multi-Agent via tmux [17:09–19:43] + + For multi-agent setups before native Discord channel isolation exists, Artem uses a tmux-based orchestrator: + - One main agent listens on Discord + - It spawns new tmux workspaces for specific tasks (e.g., weekly planning) + - The orchestrator reads other workspaces' screens to monitor progress + - Each workspace runs its own Claude Code session with a specific skill/prompt + + ### Source Monitor Report [19:43–20:22] + + The scheduled source analysis delivers a consolidated report covering Reddit threads (with sentiment), Twitter analysis, YouTube video summaries, and Obsidian community updates — all tunable. + + ### Future Wishlist and Philosophy [20:22–22:22] + + Missing features: more platform connectors (WhatsApp, Slack), native multi-agent Discord channels, better observability during agent work. The core thesis: Obsidian as the context layer is what makes the agent personal. Skills are packaged work sessions, conversation history accumulates corrections, and the agent improves over time. Recent additions (loop, dispatch, channels) complete the picture for building "your own OpenClaw with Claude Code." + + --- + + ## Connections + + - Directly relevant to this vault's architecture — similar Obsidian + Claude Code + skills setup + - [[Sessions/]] project files serve the same role as Artem's "maps world" context + - The loop/cron pattern maps to the `/loop` skill available here + - Telegram channel plugin is already configured in this vault's MCP setup + + --- + + ## My Notes + + *Space for your own thoughts, reactions, and connections after watching* + + --- + + ## Full Transcript + + <details> + <summary>Click to expand transcript</summary> + + [0:00] Antropic just shipped channels a long + [0:02] waited feature for me and as I run cloud + [0:05] code from absinian world I have 100 plus + [0:07] skills I have extensive cloud.md and + [0:10] memory and the one thing I couldn't do + [0:13] is to reach it from my phone and now + [0:15] with the channels I can I'm going to + [0:17] show you a demo of what's possible with + [0:19] the channels I'll walk you through what + [0:21] is the difference between open claw and + [0:23] clot code plus channels what is the role + [0:25] of obsidian within all of this personal + [0:28] AI agents infrastructure structure, how + [0:30] to enable proactive agents with a loop + [0:33] pattern and how to set up it yourself + [0:35] through Telegram and Discord. And we'll + [0:37] take a look at why Discord is + [0:39] interesting as a platform for agents. + [0:41] How does the future looks like? Let's + [0:43] get started. On the left, I have my + [0:45] cloud code session which is listening + [0:46] for messages from my Discord channel. + [0:49] And here I can talk to to it and it's + [0:52] going to receive my messages. So, let's + [0:54] just demonstrate that. Hey, how are you + [0:56] doing now? So we can send those messages + [0:58] and you can see that it it's appearing + [0:59] right here. The key feature is that we + [1:01] are running it from our obsidian vault + [1:03] where we have all of our skills stored. + [1:06] Now we can do some cool stuff here. I + [1:09] want you to generate a diagram to + [1:12] understand how the channels work and + [1:14] attach it to the message in my Discord. + [1:17] Okay. So the agent received our message. + [1:20] So you can see it's working + [1:22] simultaneously. So it reads my slides + [1:25] and Okay. So it understands what I'm + [1:27] working on and it generates a diagram. + [1:29] Now let's also try to create this uh PDF + [1:33] file with my slides and attach it to my + [1:35] discord. So this way this way we can see + [1:38] that actually the agent has ability to + [1:41] use our skills and it also like for + [1:44] example here excro visuals for the + [1:47] diagram and now we added a new task to + [1:50] generate PDF. Okay. So the diagram is + [1:52] finished. Uh now let's wait for the + [1:55] response. All right. And here's our + [1:57] diagram which was generated just now. So + [2:00] you can see uh here's our telegram. We + [2:02] send a message and it's being processed + [2:04] by the channels plugin. And then + [2:06] cloudcut receives this message and has + [2:08] access to our skills memory and cloudmd. + [2:11] Okay. So we got our PDF generated. Now + [2:13] clot would replies back to with PDFs and + [2:17] we can actually see it right here. So + [2:20] here is our PDF with the slides now and + [2:23] you can see how powerful this is. What + [2:26] we just did is we were communicating + [2:28] through Discord and we're running a clo + [2:31] code inside of our Obsidian vault which + [2:33] has all of this context about oursel + [2:35] about our skills memory and now we can + [2:38] access all this power uh on the go from + [2:41] our phone which is just incredible + [2:43] unlock. And the way it worked is we + [2:46] first need to start our cloud code + [2:48] within my folder here. I use maps world + [2:51] which has all of the context and then I + [2:53] use this command channels and I load + [2:56] this plugin which which is basically MCP + [2:59] server that enables clot to talk to my + [3:02] discord and I use this uh dangerous + [3:04] escape permissions flag. So ideally we + [3:06] should not use it but many people like + [3:09] to do so to avoid um notifications. And + [3:12] now it's going to listen for the + [3:14] messages in our discord. Right. So let's + [3:17] talk about why obsidian matters and why + [3:20] is it enabling. Obsidian is essentially + [3:23] a markdown viewer which is right now I'm + [3:26] opened right now and I can see my notes + [3:30] and my files on my computer in this + [3:32] format and I have and we have this + [3:35] concept of vault which is basically a + [3:37] folder on your computer and we can use + [3:40] we can run cloud code there where we're + [3:42] going to have all of the skills in + [3:45] docloud folder right here now and when + [3:48] you run this cloud code within our + [3:50] obsidian involved. We are getting the + [3:52] full access uh to all the skills which + [3:55] you have as a memory file. Here's + [3:58] example of how memory file looks for me + [4:01] that okay it's a personal assistant for + [4:03] my obsidian vault. Uh it explains who I + [4:06] am. Uh here's also some like outline + [4:09] about like different like guard rails + [4:11] before execution. Um how to transfer + [4:15] contacts within agents. Uh file + [4:17] organization for my Obsidian vault. how + [4:20] to search across my notes, my writing + [4:23] voice on how I write. It has all this + [4:25] context about my myself which makes this + [4:28] which makes this incredibly powerful and + [4:30] you have all the skills as well. Um so I + [4:32] have like 150 skills + [4:35] within this vault and you can do so + [4:37] much. So let's demonstrate that. So + [4:39] let's say I want you to use Reddit red + [4:43] skill to um look at the sentiment in + [4:48] cloud code subreddit. So what's going on + [4:51] there? Extract the posts um and create a + [4:55] report from me with links so I can visit + [4:56] them and go through them on my own. Also + [5:00] give me the number of upwards and exact + [5:04] citations from from the Reddit. So we + [5:06] can give those kind of tasks with the + [5:09] skills which we already have within our + [5:10] vault. Okay, it's right now extracting + [5:12] all of the Reddit posts at them. Okay, + [5:16] it's been crunching the data and ready + [5:18] for the last 4 minutes and finally we + [5:21] got a report. So let's see what we got. + [5:24] That will be quite interesting to see. + [5:26] So it's replying right now. + [5:29] I see that's not what I want. I wonder + [5:32] if you cannot touch it with the link. Uh + [5:35] so I can easily scroll and get access to + [5:39] the links from Discord. Okay, let's send + [5:41] that. Okay, so the results are back and + [5:45] it's actually that's a markdown file + [5:47] just in chat inline chat message please. + [5:50] Now we need to be very patient with + [5:51] those agents to teach them the way we + [5:55] want results to to be produced. Okay, so + [5:59] that's what I want. That's what I want. + [6:01] exact Reddit threads with analysis and + [6:04] quotes from it. This is what I want. + [6:06] That's exciting. Now that's very + [6:08] exciting. And this is real threads. You + [6:10] can go to it and there you go. Now you + [6:13] can adjust the skill uh to make cloud + [6:16] respond in this way with the inline + [6:18] links. That's kind of example of uh + [6:21] what's possible. Okay. So now let's + [6:23] compare that to open claw. What you have + [6:25] so far is for open claw is much more + [6:28] mature in terms of platform support. It + [6:31] has like 23 plus platforms has WhatsApp + [6:36] signal all of the different channels and + [6:38] for now what we have in cloud code is + [6:41] telegram and discord. There are some + [6:42] features which are built in in openclaw + [6:45] such as heartbeats. Uh for cloud code we + [6:47] can do the same way uh by using loop and + [6:50] I'm going to show you that next. for + [6:52] open claw there is a skills marketplace + [6:55] and there's a bunch of pre-installed + [6:56] skills which is I don't really like that + [6:58] approach um what I typically do is I do + [7:01] some some work and then I package + [7:03] something as a skill which I understand + [7:06] which I can open in my obsidian I can + [7:09] look through it I can tune it which + [7:11] brings to me much more observability now + [7:13] for open cloud we also have a memory + [7:15] which is soul.md um whereas for cloud + [7:18] code and obsidian we have this cloud md + [7:21] which I showed And we have like all this + [7:22] context from our obsidian world. And + [7:25] this is what I really like about clot + [7:27] code plus obsidian is that I can open + [7:29] any file any note. I can see exactly + [7:32] what's happening in the skill what's + [7:34] written there. And it gives me like this + [7:36] sense of control that okay I can read + [7:38] those files. I can see what agent would + [7:40] be doing. Now in terms of pricing for + [7:43] open claw you have to use API keys. uh + [7:45] or you can use OpenAI subscription and + [7:48] you want to have access to OpenAI + [7:50] models. Uh the problem with OpenAI + [7:53] models that their their vibes are just + [7:55] not great. They're not really great for + [7:59] personal assistant. They're not as + [8:01] easygoing as uh anthropic models. And + [8:04] that's kind of the problem. So you have + [8:05] to in some way use uh anthropic models + [8:08] to get a real good experience. And now + [8:12] with the channels and cloth code. So + [8:15] what we are paying is a subscription. We + [8:17] can pick subscription level from $20, + [8:20] $200 to $200. And we get access to the + [8:23] state-of-the-art models. I use OPUS. And + [8:26] what I see right now is that the usage + [8:29] which you got from this $200 is is just + [8:32] incredible. The amount of usage is + [8:35] approximately like I estimated to a + [8:37] couple thousand of dollars. essentially + [8:40] you pay $200 per month but you burn in + [8:44] tokens like a couple grants easily and + [8:47] that's right now heavily subsidized. So + [8:49] we need to remember that. So in the + [8:51] future I believe this issue needs to be + [8:54] solved somehow and the usage I believe + [8:57] wouldn't be as generous as we are right + [8:59] now unless there's going to be some + [9:00] breakthrough in terms of models which is + [9:02] going to have great performance and be + [9:04] very very cheap. Now one of the feature + [9:06] is which is great for open claw is + [9:09] having those hard bits and we can we can + [9:13] get something similar very very similar + [9:15] as we have with open claw loop skill. So + [9:19] the way it works is that we have this + [9:21] built-in loop skill where we can tell + [9:24] the interval let's say hours or minutes + [9:27] and then we can tell the prompt or we + [9:30] can tell to run some specific skill. So + [9:33] let's say okay we can set up a therapy + [9:36] style chicken. Uh let's try to do this. + [9:38] Um okay so that's going to be the + [9:42] example of the skill. And another one is + [9:45] we can do yeah I want to also stop um + [9:49] source analysis from my automation skill + [9:52] which is getting the information from + [9:54] Reddit, Twitter and YouTube. Now that's + [9:58] the first automation. Um, right now we + [10:01] set up a job for every 2 hours just to + [10:05] check in how's it going and it's going + [10:07] to ask us every 2 hours about our state + [10:09] or condition. That's kind of the + [10:11] therapist style interaction and for this + [10:14] automation I just want to run it and set + [10:17] up it in the current workspace and let's + [10:19] run it for the Reddit analysis and let's + [10:21] say you want to run it every 8 hours and + [10:23] with that you can be very very creative. + [10:26] Uh you can do okay some morning brief uh + [10:28] every 24 hours at like 7:00 a.m. scan of + [10:31] my latest notes extract ideas from + [10:33] previous day and send me morning brief + [10:36] using to telegram or discord using this + [10:39] loop skill. Okay, we are setting up uh + [10:42] this chron job for analysis of of + [10:45] different uh sources. Right now the + [10:47] agent is running the first scan and + [10:49] let's say we're going to we want this + [10:52] report delivered to our discord in a + [10:54] similar way as we had for radi analysis + [10:58] of the threats. Okay. So while the agent + [11:00] is working we can try to set it up. + [11:03] Let's go through the installation + [11:04] procedure. I'll just open a new + [11:06] terminal. And so what we need to make + [11:09] sure is is that our cloud code version + [11:12] is updated to the latest one. So it + [11:15] should be at least 80 to run this. Okay. + [11:19] Now let's install a plugin for the + [11:20] telegram. Okay. So we installed the + [11:22] plugin and now we can run cloud code. So + [11:26] we have this plugin where we can + [11:28] configure our telegram and man manage + [11:30] the telegram channel access. Now the + [11:32] next steps include going to the bot + [11:35] father and create a new bot. Okay. So + [11:38] here I am in the bot father and I'm + [11:40] creating a new bot. Okay. So we are + [11:42] good. we receive this uh access token + [11:44] and now we need to run this command + [11:46] configure and here is our token. The + [11:50] token has been saved and now we can + [11:52] restart cloud code uh with this channels + [11:56] flag and I'm going to use this telegram + [11:59] plugin. Okay, so now we're listening + [12:01] from for the incoming messages. Now I'm + [12:03] going to my bot. I'm starting it and now + [12:06] we can see that uh right now they're + [12:08] paired and I can just ask something. Uh + [12:11] hey, how's it going? Um, what is the + [12:13] context do you have? What do you know + [12:14] about me? Oh, okay. So, you see here, + [12:17] that's kind of the reason why we need to + [12:19] run with dangerous escape permissions to + [12:22] avoid this um commands. Okay. Can you + [12:25] tell me about the skills we should have + [12:27] now? And and you see, so that's been + [12:30] just a 2 minutes for setup and we + [12:33] already can start talking to our + [12:35] obsidian world. It has access to our + [12:37] skills and just ready to go. So, it's + [12:39] it's been very very smooth. uh so we can + [12:42] add maybe avatar for our bot. Now + [12:45] another thing what we need to do is to + [12:47] enable this access policy from the allow + [12:50] list. So we can only us who are talking + [12:53] to this bot. Now let's set this up for + [12:56] our bot and I want you to add me as the + [13:01] one who can only talk. Okay. So now + [13:04] we've changed the setting so we are + [13:07] good. Now, that's been Telegram and I + [13:09] think Telegram is kind of cool. Um, but + [13:11] what I really love is Discord and I + [13:14] think Discord is the future. Now, I've + [13:16] seen a person from Twitter who was doing + [13:19] something like that with his his open + [13:23] claw set up which was quite quite + [13:26] impressive to be honest. And I want to + [13:29] set up something similar for myself in + [13:31] Discord where I'm going to have agents + [13:34] with isolated context. Let's say okay so + [13:38] here we have a general chat here we have + [13:40] our pulse where we get those reports + [13:43] maybe from Reddit here are some logs and + [13:46] so on so you can have a channel based + [13:50] scope for agents with their own context + [13:53] right now they're not there yet um it's + [13:55] still emerging but I think that's going + [13:57] to get there like very very quickly so + [14:00] as we figure out how to properly set up + [14:04] Discord. So for now, let's do something + [14:06] simple. So we're going to just have our + [14:08] own server where we're going to have a a + [14:10] single cloud code called session running + [14:13] and attached with our scene involved + [14:15] context. Okay. So let's install the + [14:18] discord plugin. Okay. So we are + [14:20] installing the plugin. It's downloaded. + [14:22] That's great. Now we need to create a + [14:25] new application. We go to this + [14:27] developers link and then okay let's try + [14:30] new application. Let's name it cloud + [14:32] test. That's good. Okay. We are not a + [14:35] robot. So we are human. Absolutely. + [14:37] Okay. And here at this point, what we + [14:40] need to do is to go to b then reset + [14:44] token. Okay. Then copy this token. We're + [14:46] going to need that. Okay. So now in our + [14:49] cloud code, what we're going to do is um + [14:53] we're going to set up this token access. + [14:55] Let's say okay discord configure. And + [14:58] then here's our token. That's our access + [15:00] token. Okay. So unstab command yes that + [15:04] should be for discord not for telegram + [15:06] while it's working let's set up another + [15:09] piece which is here the authentification + [15:13] and we go to the URL generator we look + [15:16] at the bot we look at the board and then + [15:20] here are the b permissions which we can + [15:22] set so we wanted to set send messages + [15:26] attach files read message history and + [15:29] all the actions. Now here we got a URL + [15:32] and then let's go to Discord and create + [15:35] a new server. So I'll just skip that. + [15:38] Okay, very good. Here's our new Discord + [15:40] server. And then let's just create a new + [15:43] tab and let's run this link. Now in this + [15:46] app which we just created, we want to + [15:48] install it to let's say our server which + [15:51] we just created. Now we have this + [15:53] permissions which we authorized and we + [15:55] are good. We're actually good. And we + [15:57] added this bot. That's our cloud. Now + [16:01] let's do something exciting. Now let's + [16:04] add this relay. And now let's run this + [16:07] new cloud session. Okay. So now we we + [16:09] are getting back to our phone. So okay, + [16:12] here's a new server which we have. And + [16:15] let's say okay, we first message the bot + [16:18] just let's say hey. Okay, so the board + [16:20] is not answering yet. Okay, so one of + [16:22] the settings which I forget to turn on + [16:25] is this one. I've been trying to set it + [16:28] up and now I think that was the issue. + [16:30] So let's get back and restart the code + [16:32] session and now we can try to message + [16:34] our bot. So hey oh okay okay so I think + [16:38] we're good. I think we are finally good. + [16:40] So we're getting this replies in our + [16:43] discord. Okay that's great. So in this + [16:46] piece was not really needed. And now in + [16:49] the discord so we can chat in the same + [16:52] way as in telegram. So let's say what + [16:55] are the skills which you have right now? + [16:57] What are the skills that you have right + [16:58] now? Okay. So and we are getting uh same + [17:02] responses. So the agent right now is + [17:05] context aware but our obsidian vault has + [17:07] access to it. And that's been the whole + [17:09] setup. And what I'm really looking for + [17:12] is extensions of this approach where we + [17:15] just have a single agent to bring it to + [17:17] to have multiple agents within our + [17:19] discord in a similar way as you would + [17:22] have in open claw. That would be huge in + [17:24] lock right now and this can be done with + [17:26] different type of of plug-in. So we just + [17:28] need to modify I think community would + [17:30] come up as something that would enable + [17:32] this type of the workflows and as + [17:34] alternative what we can use is this + [17:36] orchestrator pattern which I showed in + [17:38] my previous video where we had one + [17:41] orchestrator which is working and we can + [17:43] talk to it through discord. Let's say + [17:45] that that's going to be our orchestrator + [17:48] agent. Let's say and just rename it. + [17:51] Let's put it here. Okay, now we are good + [17:54] and we can use this terminal. What I'm + [17:56] using right now is a Cmax terminal where + [17:59] I can spawn different workspaces and I + [18:02] have this Cmax skill which enables cloud + [18:05] code to create uh different workspaces. + [18:08] So let's try to demo this. Okay, so I + [18:10] want you to use uh Cmax skill to spawn a + [18:16] new workspace. let's say for my u + [18:19] planning but actually before doing that + [18:21] let's just list all of our workspaces + [18:23] which we have and send me them to + [18:26] discord and then proceed with the task + [18:28] that's going to be our orchestrator + [18:30] agent it uses this uh CLI tool called + [18:33] cmax to programmatically access this + [18:36] terminal it replies to me that we have + [18:39] this orchestrator workspace now it's + [18:42] going to create a new workspace called + [18:44] weekly plan and it adds a pro prompt as + [18:47] It's essentially a skill which we have + [18:49] for weekly planning. Now it runs runs it + [18:52] and it's going to execute this weekly + [18:54] planning skill. Now the beauty here is + [18:57] that okay we have this orchestrator + [18:59] agent I can talk to and I can ask about + [19:02] what is this current state of the weekly + [19:04] plan what what the other agent is doing. + [19:07] This main orchestrator agent is going to + [19:09] talk to another guy which is weekly plan + [19:12] and it reads the current screen of + [19:14] what's happening in that workspace and + [19:17] this way orchestrator can manage and see + [19:20] what's happening in different workspaces + [19:22] and you can spawn those workspaces uh as + [19:25] much as you want and you can see right + [19:28] now so it's finding the recent week plan + [19:32] checking the calendar for the week is + [19:33] done check active sessions is done and + [19:36] This is a pattern which you can use + [19:38] while we we are waiting for the complete + [19:40] support of different isolated workspaces + [19:43] in discord. Okay. And meanwhile data + [19:46] source monitor report in from another + [19:48] agent. Uh here is a Reddit. Uh here is a + [19:51] analysis of sentiment. Uh here is a + [19:54] different submar about Obsidian. What's + [19:56] going on? Here is a Reddit for the + [19:59] immigration. Here's a Twitter analysis. + [20:02] Here's a YouTube different YouTube + [20:04] videos. And those one you can tune you + [20:06] can actually tune those for example I + [20:08] don't really like that it just gives me + [20:09] a link I want also present me the + [20:11] summary and also it reports to me a + [20:14] change log for the recent uh cloud + [20:16] quarter upgrade here all of the videos + [20:19] and that's been the report which we run + [20:22] uh initially now what I want to see is + [20:25] to have more platforms having easily to + [20:29] connected to WhatsApp or Slack be able + [20:31] to run multi- aent setup natively within + [20:34] Discord as would be a huge unlock and + [20:36] another piece which is concerning to me + [20:39] is this observability that when agent is + [20:42] working I don't really see what's + [20:43] happening in here it's very very + [20:45] challenging for me to observe what I see + [20:47] only is the end result in my terminal + [20:50] but I don't really see how agent is + [20:52] crashing the data and I I don't really + [20:54] see if agent just went off track + [20:56] completely and I can't control it I + [20:58] can't stop it I just have to guess and + [21:00] maybe one of the solution is to tell + [21:03] agent to like uses reply command uh from + [21:06] this MCP to notify us before doing + [21:09] something. I think that's can be state + [21:11] stated in in the rules. Uh this would be + [21:14] one of the solutions now and the most + [21:16] important thing is having your context. + [21:18] I believe the obsidian is unlock here in + [21:21] the setup like your preferences, your + [21:23] skills, your playbooks. Without that, + [21:25] the agent is generic, right? It can be + [21:27] personal by definition. So the skills + [21:30] are playbooks that can handles the work + [21:33] that you've done. You run a session and + [21:35] then at the end of the session, you've + [21:36] done some work. Now you want to package + [21:38] it as a skill. That's going to be your + [21:40] playbook which you can run whenever you + [21:42] want. And because everything runs + [21:45] locally here, you can have access to + [21:47] your whole conversation history, every + [21:49] session, every decision, every + [21:50] correction. And this makes the agent + [21:53] better and better over time. You can ask + [21:55] cloud to analyze your previous + [21:56] conversations from the last week and ask + [21:58] it to improve your claw.md or the + [22:01] skills. And I think that's the final + [22:03] piece like last week we had a loop a new + [22:06] skill cuz two days ago we had a dispatch + [22:09] and now we have a channels. And I think + [22:10] that's kind of the full piece to build + [22:12] your own open cloud with cloud code. + [22:14] It's really exciting times uh to be + [22:17] alive uh honestly. And I'll see you in + [22:20] the next one. Yeah. Thank you for being + [22:22] here. + + </details> + ``` Now let me log the activity and open the note. --- ## 19:16:27 - Assistant ### 🔧 Skill **Status**: completed --- ## 19:16:28 - Assistant ### 🔧 Skill **Status**: completed --- ## 19:16:31 - Assistant ### 🔧 date +%H:%M && date +%Y-%m-%d **Status**: completed --- ## 19:16:32 - Assistant ### 🔧 a="/Applic"; a="${a}ations/Obsidian.app/Contents/MacOS/Obsidian"; "$a" open path="Videos/Claude Code Channels Replaced My OpenClaw Setup.md" **Status**: completed --- ## 19:16:37 - Assistant ### 🔧 Read Daily Notes/2026-03-24.md **Locations**: `/Users/nickyoung/My Obsidian Vault/Daily Notes/2026-03-24.md:1` **Status**: completed --- ## 19:16:40 - Assistant ### 🔧 Edit Daily Notes/2026-03-24.md **Locations**: `/Users/nickyoung/My Obsidian Vault/Daily Notes/2026-03-24.md:35` **Status**: completed **File**: `/Users/nickyoung/My Obsidian Vault/Daily Notes/2026-03-24.md` ```diff - - - 08:53 - Processed video: "Claude Code MASSIVE Update!" → [[Claude Code MASSIVE Update - Claude Code OS Computer Use Schedule and More]] - - 14:57 - Added [[HM Passport Office Chat Transcript - 2026-03-24.pdf]] to Attachments/ and updated passport tracking notes with DHL delivery + HMPO webchat outcome - - --- - - 14:17 - Created [[Section 3 restructuring - checkpoint 24 Mar]] in Notes/ + + - 08:53 - Processed video: "Claude Code MASSIVE Update!" → [[Claude Code MASSIVE Update - Claude Code OS Computer Use Schedule and More]] + - 14:57 - Added [[HM Passport Office Chat Transcript - 2026-03-24.pdf]] to Attachments/ and updated passport tracking notes with DHL delivery + HMPO webchat outcome + - 19:16 - Processed video: "Claude Code Channels Replaced My OpenClaw Setup" → [[Claude Code Channels Replaced My OpenClaw Setup]] + + --- + - 14:17 - Created [[Section 3 restructuring - checkpoint 24 Mar]] in Notes/ ``` Note created and opened. - File: `Videos/Claude Code Channels Replaced My OpenClaw Setup.md` - Channel: Artem Zhutov - Transcript: available (full, with timestamps) - Covers: Claude Code channels feature (Telegram/Discord), comparison with OpenClaw, loop pattern for proactive agents, tmux orchestrator pattern, setup walkthroughs Quite similar to your own setup — he's running Claude Code from an Obsidian vault with 150+ skills and CLAUDE.md context, using the channels MCP plugin for mobile access. The orchestrator-via-tmux pattern and the observability concerns he raises are worth noting. *Un altro utente di Obsidian che costruisce il proprio assistente personale con Claude Code — la convergenza delle architetture è notevole.* ---