PeekaTube Icon

秒抓 YouTube影片重點
速度超有感

任何 CC 影片，AI 都能秒速幫你畫重點。從此告別廢片標題黨，學習力直接開外掛，效率飆升 10 倍！

逃離內容農場免費新增到 Chrome
https://www.youtube.com
25 Hidden n8n Features That Save Hours of Work
N

25 Hidden n8n Features That Save Hours of Work

Nate Herk | AI Automation

2 小時前

Summary

這部影片介紹了 25 個 Naden 的實用技巧，旨在提升開發者的生產力。這些技巧從基礎概念到進階應用，涵蓋了資料處理、工作流程管理、除錯以及與外部服務的整合等多個方面。影片強調了這些技巧能幫助使用者更有效率地建構和維護 Naden 工作流程。

Notes

  • 固定值與表達式轉換： 在輸入框中，在輸入任何內容前，直接按下鍵盤上的等號 (=) 即可將固定值切換為表達式。
  • 固定 (Pin) 資料： 對於會產生費用或執行時間長的節點，可以將其輸出資料固定，避免重複執行 API 調用，節省成本和時間。
  • 模擬資料 (Mock Data)： 在節點或觸發器中，可以手動填入模擬資料，方便測試工作流程的後續部分，尤其適用於與其他工作流程觸發的場景。
  • 節點釘選快速鍵： 選中節點後，按下鍵盤上的 'P' 鍵可以快速釘選或取消釘選資料。
  • 節點間快速切換： 在節點設定面板中，將滑鼠移到節點的左右邊緣，即可直接切換到相鄰節點的設定，無需退出。
  • 使用方向鍵和 Enter 鍵導航節點： 在工作流程視圖中，可以使用方向鍵在節點間移動，並按下 Enter 鍵打開選中的節點。
  • 停用 (Deactivate) 節點： 可以透過點擊節點的選單或選中節點後按下 'D' 鍵來停用節點，這在測試工作流程時，用於排除錯誤很有幫助。
  • 組織化技巧：
    • 便利貼 (Sticky Notes)： 使用便利貼來標註工作流程的說明或指南，支援 Markdown 格式，並可自訂顏色以區分不同類型的節點。
    • 命名節點： 快速命名節點，可以透過點擊節點並按下 F2 鍵來達成。
  • 工作流程歷史記錄： 可以查看和還原工作流程的歷史版本，類似於 Google 文件。
  • 快速新增節點： 在工作流程中，按下 Tab 鍵即可快速調出新增節點的選項。
  • 禁用歸因 (Attribution) 訊息： 在發送電子郵件、Telegram 或 Slack 訊息等整合節點時，可以禁用「訊息由 Naden 發送」的歸因文字。
  • 獲取當前日期和時間： 在表達式中，使用 {{ $now }} 函數可以獲取當前日期和時間，並可使用 .format 方法進行格式化。
  • 儲存憑證 (Credentials)： 在進行 API 調用時，可以儲存 API 金鑰等憑證，避免每次手動輸入，並在分享模板時不會暴露個人憑證。
  • 錯誤工作流程 (Error Workflows)： 使用 Error Trigger 節點，可以在主工作流程發生錯誤時觸發通知，方便及時處理。
  • 工作流程變數： 可以獲取執行 ID、工作流程名稱等變數資訊。
  • 指定子工作流程輸入： 在將資料傳遞給子工作流程時，可以預先定義接收的欄位，使資料傳遞更清晰。
  • 檢視子執行資料： 可以查看子工作流程的具體執行情況，以便診斷問題。
  • 代理程式日誌 (Agent Logs)： 查看代理程式的內部日誌，了解其處理過程和與工具的互動。
  • 複製到編輯器： 在檢視子執行資料時，可以將其複製到編輯器中進行進一步的測試和調試。
  • 人工審核 (Human in the Loop) 節點： 可以在工作流程中加入需要人工審核或輸入的節點，支援 Discord、Slack 等多種溝通方式，並可設定為按鈕、自訂表單或自由文字輸入。
  • 標準化輸入： 確保來自不同來源（如 Telegram、Naden Chat）的輸入能夠以統一的格式傳遞給代理程式。
  • 錯誤時繼續 (Continue on Error)： 在節點設定中，可以選擇在發生錯誤時不中斷整個工作流程，而是繼續執行，並可選擇將錯誤訊息導向另一分支進行處理。
  • 節點設定的多樣性： 不同節點具有不同的設定選項，例如 SSL 證書、重複執行次數、重試機制、複製項目等，可根據需求進行調整。
  • 結構化輸出： 代理程式可以輸出結構化的資料，並透過結構化輸出解析器 (Structured Output Parser) 進行解析，以便個別處理。
  • 陣列操作 (Join)： 對於陣列資料，可以使用 .join() 方法自訂分隔符來合併陣列中的元素。
  • 替換功能 (Replace)： 可以用來替換字串中的特定內容，例如將 JSON 體內的雙引號替換為其他字元，以確保資料格式正確。

Quotes

  • "All right, here are 25 Naden hacks that I wish I knew when I started building with Naden last year."
  • "And hopefully these are going to help your productivity explode."
  • "But what you can do is if something's fixed and before you start typing in there, you can just hit the equal sign on your keyboard and it will switch to an expression."
  • "This is really helpful for any nodes that are going to cost you money every time you run them or if they take a long time to run."
  • "So, this is very similar to pinning data where I could now say we're going to have a field called query."
  • "If you move your mouse to the right end or the left end, you can basically just switch between your node configurations without ever having to leave one."
  • "And that's like basically just the keyboard shortcut to deactivate stuff."
  • "So whenever I'm trying to create a sticky note like those setup guides, I have chat GBT create those for me."
  • "Same way on a Google doc or Google sheet, you can see previous versions and you can restore those if you want."
  • "When you're sending an email or a telegram message or a slack or any of these different integrations, a lot of times at the bottom it will say this message was sent by nitn."
  • "All you have to do is make sure you're in an expression and then you're going to do two curly braces and then you'll see right here is a dollar sign now function."
  • "So, why would I want to go copy that API key every time I was setting up this HTTP request? I don't want to."
  • "This way you can send yourself email notifications, text notifications, whatever you want if one of your main workflow errors."
  • "So, what you'll do is you'll click into that actual trigger."
  • "So number 15 is looking at subexecution data."
  • "And then finally, when it comes to viewing the subexecution data, you can copy stuff to the editor."
  • "We have human in the loop nodes, which basically says wait for approval or human input before continuing."
  • "The idea here is to standardize the input so you can have multiple forms of communication with an agent."
  • "And so the reason this didn't stop the whole workflow is because in this node in the settings, we're telling it to continue on error rather than stopping the entire workflow."
  • "So, what we have right here is three URLs within this um code node."
  • "And then last but not least, we have the ability to replace things."
  • "So, those were 25 Nin hacks that are hopefully going to increase your productivity."
Deepseek V3.1 Code: This REALLY THROWS Claude Code into the TRASH! (+V3.1 Test Results)
A

Deepseek V3.1 Code: This REALLY THROWS Claude Code into the TRASH! (+V3.1 Test Results)

AICodeKing

5 天前

Summary

這段影片介紹了 Deepseek V3.1 模型，強調其作為一個混合推理模型，能夠同時處理推理和非推理任務，從而提高效率。影片指出 V3.1 在工具調用（tool calling）方面有顯著改進，解決了先前版本的痛點，並在多項基準測試中表現出色，包括 Clin Rue 和 Open Code。此外，模型的上下文長度已從 65K 擴展到 128K。影片展示了 V3.1 在生成圖像（如 SVG Panda、autoplay chess、flying butterfly）方面的能力，認為其生成結果令人印象深刻，甚至能與 Opus 媲美，但價格卻便宜許多。對於編碼任務，影片建議使用非推理變體，認為其表現優於推理變體。影片還提到 V3.1 的 API 格式與 Anthropic 兼容，使其能輕鬆替換 Sonnet 模型，且價格便宜 20 倍。影片的後半部分演示瞭如何將 Deepseek V3.1 與 Claude Code 整合，利用 Requesty 等工具優化設置，並展示了其在構建應用程式（如電影追蹤器）時的強大效能，特別是在工具調用方面，能夠順利完成任務，且速度和價格都具有顯著優勢。

Notes

  • Deepseek V3.1 是一個混合推理模型，能處理推理和非推理任務，提升效率。
  • V3.1 在工具調用（tool calling）方面有大幅改進，解決了 R1 和 V3 版本的主要痛點。
  • 模型在 Clin Rue 和 Open Code 等基準測試中表現更佳。
  • 上下文長度從 65K 提升至 128K。
  • 模型端點與之前相同：deepseek chat 為通用模型，reasoning 為推理變體。
  • 在影片的基準測試中，V3.1 的性能得到顯著提升，即使是非推理任務也表現優異。
  • V3.1 在生成地板平面圖方面表現不錯，且對於如此便宜的模型來說，單次生成（oneshot generation）的表現非常出色。
  • SVG Panda 圖像（帶有漢堡包）是該模型迄今為止最好的作品之一，甚至優於 Opus。
  • Autoplay chess 功能也運行良好，儘管不總是產生合法棋步，但仍優於許多其他生成。
  • 飛行蝴蝶圖像也相當酷炫，總體而言這些圖像生成效果很好。
  • 對於編碼任務，非推理變體的表現比推理變體更好。
  • V3.1 在 ADER leaderboards 上表現接近 Opus，但價格便宜得多。
  • Deepseek V3.1 被認為是 Sonnet 的絕佳替代品。
  • Deepseek 提供了 Anthropic API 格式，方便作為替代品使用。
  • 影片演示瞭如何使用 Claude Code 整合 Deepseek V3.1。
  • 建議使用 Requesty 作為 Open Router 的替代方案，以獲得更好的配置和自定義選項。
  • Requesty 支持 Anthropic 兼容格式，可以輕鬆使用 Deepseek 模型。
  • 演示瞭如何在 Claude Code 中設置 Deepseek 的基礎 URL 和模型。
  • V3.1 的斜槓命令 (slash commands) 和 MCPs（如 bite rover MCP）均運行良好。
  • MCPs 允許在不同編碼器之間共享記憶體，適用於項目規則和記憶體。
  • 影片展示了如何在 Next.js 專案中使用 V3.1 創建電影追蹤器應用程式。
  • V3.1 在工具調用方面表現出色，能夠順利完成任務，這是前一版本的一大缺點。
  • V3.1 的速度有所提升，可能與其只需部署單一模型有關。
  • V3.1 的整體表現與 Sonnet 相當，但價格便宜 20 倍。
  • 影片推薦 V3.1，認為它解決了原始模型的許多痛點，並在編碼方面表現更好。
  • 對於數學等任務，可能需要推理變體，但對於編碼，非推理變體已足夠優秀。
  • V3.1 提供了更經濟的替代方案，無需為簡單任務使用 Sonnet，節省了大量成本。

Quotes

  • "So, Deepseek V3.1 was launched, but the major info about it wasn't available before."
  • "the TLDDR is that it is a hybrid reasoning model that can basically do both reasoning and non-reasoning tasks."
  • "This allows for efficient performance."
  • "It also now has much better tool calling support which was one of the major pain points of Deepseek R1 and V3."
  • "It now scores way better in multiple tool calling benchmarks and works much better in things like clin rue and open code."
  • "Similarly, they have also increased the context limit to 128K from 65K which is a great upgrade as well."
  • "The oneshot generation that just works for a model this cheap is awesome."
  • "The SVG Panda with the burger is one of the best I have seen yet. It is actually usable, which is not even something that Opus can make."
  • "The autoplay chess also works pretty well. It doesn't always make legal moves, but it is still better than many other generations."
  • "And overall this stuff is really good."
  • "It comes close to opus for way cheaper pricing which is quite awesome for an open and cheaper model."
  • "It feels a lot like Sonnet but it's 20x cheaper."
  • "So yeah, it's awesome. It can also use the to-do list stuff as well as MCPs without any issues."
  • "It looks pretty great. It's on par with what Sonnet makes in one shot while being a lot cheaper and pretty similar."
  • "I was thinking Deepseek V3 3.1 was not as good at first, but as I used it, it felt like Deepseek actually addressed almost all the major pain points of the original model, like tool calling, MCP support, and everything."
  • "It now works way better and is also way better at coding."
  • "So yeah, I'd highly recommend you check this out. It's an awesome model."
  • "So yeah, it's pretty usable now."
  • "This makes a lot of sense to use. I prefer it over GLM for replacing and clawed as it's pretty great now."
  • "there's probably no need to use sonnet for trivial tasks which saves a ton of money and costs."

你的時間，被無用內容綁架了

時間是你最寶貴的資產。YouTube 卻這樣為你揮霍它。

  • 9:00 - 9:25
    長達 25 分鐘的教學，但真正的 2 分鐘解答卻藏在影片最後。
  • 14:30 - 14:45
    15 分鐘的「評測」，只為了業配，而不是在 1 秒內給你明確答案。
  • 19:45 - 20:20
    「簡短」影片，卻讓你掉進 20 分鐘的無關內容裡。
每日總損失
75
分鐘/天
整整一個工作週，就這樣沒了。
40+
小時/月

為什麼我得看半部影片，才知道它到底有沒有用？

— 任何珍惜時間的人

有更好的方法...

在你看影片之前，提前知道內容！

1.

點擊 PeekaTube Icon，立即取得摘要

在任何有字幕的 YouTube 影片上，點擊 PeekaTube 圖示 (PeekaTube Icon)，AI 會在幾秒內為您生成摘要。
2.

無需播放，即可預覽

將滑鼠懸停在按鈕上方，即可閱讀影片的關鍵重點。無需觀看影片，就能掌握核心內容。
3.

學得更快或直接跳過

立即判斷影片是否值得您花時間觀看。快速吸收所需見解，或自信地離開影片尋找更好的內容。
體驗更好的方法（免費！）

檢視 & 輸出一次搞定

逐字稿 • 摘要
25 Hidden n8n Features That Save Hours of Work
25 Hidden n8n Features That Save Hours of Work
Nate Herk | AI Automation
English (auto-generated)
2.66s
00:00 - 00:03 All right, here are 25 Naden hacks that 00:02 - 00:04 I wish I knew when I started building 00:03 - 00:06 with Naden last year. We're going to 00:04 - 00:08 start simple and get progressively more 00:06 - 00:09 advanced. And hopefully these are going 00:08 - 00:11 to help your productivity explode. All 00:09 - 00:12 right, starting off with number one. 00:11 - 00:13 Hopefully, we all know the difference 00:12 - 00:14 between a fixed value and an expression 00:13 - 00:16 value. And what you can see right here 00:14 - 00:17 is that we have the word high and it's 00:16 - 00:19 fixed. And typically, if you wanted to 00:17 - 00:20 make this an expression, you would have 00:19 - 00:21 to come over here and hit expression. 00:20 - 00:23 But what you can do is if something's 00:21 - 00:24 fixed and before you start typing in 00:23 - 00:25 there, you can just hit the equal sign 00:24 - 00:27 on your keyboard and it will switch to 00:25 - 00:29 an expression. Okay, number two is the 00:27 - 00:30 ability to pin data. So let's say I'm 00:29 - 00:32 running this perplexity search where I'm 00:30 - 00:34 looking for what is naden. Once I got 00:32 - 00:35 this output, this costed me money. So 00:34 - 00:37 what I want to do is pin it. That way 00:35 - 00:39 when I want to run the workflow later, I 00:37 - 00:41 don't have to resubmit another call to 00:39 - 00:42 perplexity's API or any API. This is 00:41 - 00:44 really helpful for any nodes that are 00:42 - 00:46 going to cost you money every time you 00:44 - 00:47 run them or if they take a long time to 00:46 - 00:49 run because now I have this data and 00:47 - 00:50 let's say I want to map out the rest of 00:49 - 00:51 the process. Now that it's pinned, I 00:50 - 00:53 could basically and let me just steal 00:51 - 00:54 this manual trigger real quick. I could 00:53 - 00:56 basically continue to run this over and 00:54 - 00:58 over and it's not going to make a new 00:56 - 00:59 call to perplexity every time. And then 00:58 - 01:01 we're able to use all of this data over 00:59 - 01:02 here to continuously map out the 01:01 - 01:04 different variables for the rest of the 01:02 - 01:05 workflow. Number three is mock data. So 01:04 - 01:07 we have the ability to basically set 01:05 - 01:08 data within any sort of node or a 01:07 - 01:10 trigger like this where I could click on 01:08 - 01:12 this little pencil mark in the top and I 01:10 - 01:13 could basically fill this in with data 01:12 - 01:14 to test through the rest of the 01:13 - 01:16 workflow. So this is very similar to 01:14 - 01:17 pinning data where I could now say we're 01:16 - 01:18 going to have a field called query. The 01:17 - 01:20 value in there is going to be test and 01:18 - 01:22 you can see it's pinned and now we can 01:20 - 01:23 use this to map the rest of our process. 01:22 - 01:25 This is especially useful when you're 01:23 - 01:26 doing a when executed by another 01:25 - 01:28 workflow trigger because then you don't 01:26 - 01:29 have to use the main workflow to trigger 01:28 - 01:31 this subworkflow every time you want to 01:29 - 01:32 test it out. And another quick hack when 01:31 - 01:34 it comes to pinning data is you can just 01:32 - 01:36 click on the node and hit P to pin it or 01:34 - 01:37 unpin it as you can see. All right, 01:36 - 01:39 number four is the ability to move 01:37 - 01:40 between nodes. So let's say you're in 01:39 - 01:42 the configuration panel of a certain 01:40 - 01:43 node and you don't want to have to, you 01:42 - 01:45 know, click out of here and click into 01:43 - 01:46 another one every time. If you move your 01:45 - 01:48 mouse to the right end or the left end, 01:46 - 01:49 you can basically just switch between 01:48 - 01:50 your node configurations without ever 01:49 - 01:52 having to leave one. So, this will 01:50 - 01:53 hopefully save you some time. And also, 01:52 - 01:54 if you're just looking at your workflow 01:53 - 01:56 from this view, you can see that I have 01:54 - 01:57 node number one selected right now. I 01:56 - 01:59 can use the arrow keys to move between 01:57 - 02:01 them and then I can hit enter to open 01:59 - 02:02 one of them up. Okay, number five is 02:01 - 02:03 being able to deactivate nodes. So, what 02:02 - 02:05 you can do is you can click on these 02:03 - 02:07 three dots and click on activate. Or if 02:05 - 02:08 you highlight over the node, you can 02:07 - 02:10 just hit D. And that's like basically 02:08 - 02:11 just the keyboard shortcut to deactivate 02:10 - 02:12 stuff. And if you've ever been in a 02:11 - 02:13 workflow and you're trying to test it 02:12 - 02:15 out and you get a little error that 02:13 - 02:16 says, you know, resolve outstanding 02:15 - 02:18 issues before you can trigger the 02:16 - 02:19 workflow, then you just click on this 02:18 - 02:20 one, hit deactivate, and then you should 02:19 - 02:22 be able to test the rest of the 02:20 - 02:24 workflow. Number six is organization. So 02:22 - 02:25 there's two aspects here. The first one 02:24 - 02:27 is using sticky notes, which we can get 02:25 - 02:28 to by coming up here and adding a sticky 02:27 - 02:30 note like that. Or anywhere in your 02:28 - 02:32 workflow, you can just hit shift plus s, 02:30 - 02:33 and it will pop up a sticky note. And 02:32 - 02:34 once you're in that sticky note, it says 02:33 - 02:35 double click to edit me. And as you can 02:34 - 02:38 see in here, we're using markdown. So 02:35 - 02:39 you can change, you know, bold or header 02:38 - 02:40 one, header two, header three based on 02:39 - 02:42 markdown. So whenever I'm trying to 02:40 - 02:44 create a sticky note like those setup 02:42 - 02:46 guides, I have chat GBT create those for 02:44 - 02:47 me. You can even see right here it says 02:46 - 02:48 you can style with markdown. You can 02:47 - 02:49 also change the color of your sticky 02:48 - 02:51 notes just to keep your workflow 02:49 - 02:53 organized. Sometimes I like to do, you 02:51 - 02:55 know, AI nodes in blue and data 02:53 - 02:57 transformation nodes in green and notes 02:55 - 02:58 in yellow, whatever you want to do. And 02:57 - 03:00 then another aspect of organization is 02:58 - 03:01 just naming your nodes. So you could of 03:00 - 03:03 course click into them and then come up 03:01 - 03:04 here to rename your node. Or if you want 03:03 - 03:05 to be more quick about it, you just 03:04 - 03:06 click on the node and hit F2 and then 03:05 - 03:08 you'll be able to change the name right 03:06 - 03:09 there. Number seven is about workflow 03:08 - 03:10 history. Same way on a Google doc or 03:09 - 03:12 Google sheet, you can see previous 03:10 - 03:13 versions and you can restore those if 03:12 - 03:15 you want. So, all you have to do is 03:13 - 03:16 click on this little, you know, timer 03:15 - 03:18 looking thing with an arrow and it will 03:16 - 03:20 pop up different versions of this 03:18 - 03:22 workflow and then you're able to go in 03:20 - 03:24 here and restore a workflow if you want. 03:22 - 03:25 You can download it. You can clone it to 03:24 - 03:26 a new one, whatever you want to do. As 03:25 - 03:28 you can see here with the previous 03:26 - 03:29 version, I have the option to restore 03:28 - 03:30 it. All right, number eight is just 03:29 - 03:31 another quick keyboard shortcut. So, if 03:30 - 03:33 you're anywhere in your workflow and you 03:31 - 03:34 want to add another node, you can just 03:33 - 03:36 hit tab and it will pull that up so you 03:34 - 03:38 can start typing right away. Number nine 03:36 - 03:39 is with Nin attributions. So let me just 03:38 - 03:40 quickly reactivate all these nodes and 03:39 - 03:42 tell you what I mean by that. When 03:40 - 03:43 you're sending an email or a telegram 03:42 - 03:45 message or a slack or any of these 03:43 - 03:46 different integrations, a lot of times 03:45 - 03:48 at the bottom it will say this message 03:46 - 03:50 was sent by nitn. So what you want to do 03:48 - 03:53 is in this node if you add an option you 03:50 - 03:54 can click on append attribution and you 03:53 - 03:55 can just flick that off and now you 03:54 - 03:57 won't get that message. It's the exact 03:55 - 03:59 same thing in telegram where you add the 03:57 - 04:00 append and end attribution and you turn 03:59 - 04:02 that off. But in Slack it's a little 04:00 - 04:03 different where you come into here and 04:02 - 04:05 it says include link to workflow and 04:03 - 04:06 that's the one you turn off in Slack. 04:05 - 04:08 So, every integration is a little bit 04:06 - 04:09 different with their terminology, but 04:08 - 04:10 you can pretty much always turn off that 04:09 - 04:12 attribution. And by the way, did you 04:10 - 04:13 guys see how quick I activated all those 04:12 - 04:15 nodes earlier? Anyways, number 10 is 04:13 - 04:16 current date and time. So, let's say you 04:15 - 04:17 have an AI agent or any sort of node 04:16 - 04:19 where you want to give it the current 04:17 - 04:20 date and time. Or maybe you just want to 04:19 - 04:22 send over a time stamp to your Google 04:20 - 04:23 Sheets. All you have to do is make sure 04:22 - 04:24 you're in an expression and then you're 04:23 - 04:26 going to do two curly braces and then 04:24 - 04:28 you'll see right here is a dollar sign 04:26 - 04:30 now function. And now your agent or your 04:28 - 04:32 node has access to the current date and 04:30 - 04:34 time. You can also go ahead and format 04:32 - 04:35 that by typing dot format. And then you 04:34 - 04:37 can play around with how you want to see 04:35 - 04:38 the date and time. Number 11 is really 04:37 - 04:40 cool. It's the ability to save your 04:38 - 04:41 credentials when you're making API 04:40 - 04:43 calls. So, Perplexity is something I use 04:41 - 04:44 a lot. Tavly is something I use a lot. 04:43 - 04:46 Open eyes image generation is something 04:44 - 04:48 I use a lot. So, why would I want to go 04:46 - 04:50 copy that API key every time I was 04:48 - 04:51 setting up this HTTP request? I don't 04:50 - 04:53 want to. So, what you're going to do is 04:51 - 04:55 when you come in here, you can do your 04:53 - 04:56 authentication. You choose generic. If 04:55 - 04:58 you know it's a header off, you choose 04:56 - 05:00 header. If it's query off, you'll choose 04:58 - 05:01 query. Whatever type of authentication 05:00 - 05:02 it is. And then you can see it pulls up 05:01 - 05:04 all the header all that you have saved 05:02 - 05:05 in here. And it's basically the same way 05:04 - 05:07 you would set up a header off right 05:05 - 05:09 here. And you type in like authorization 05:07 - 05:10 and then bear space your API key. You 05:09 - 05:12 would pretty much just do that in here. 05:10 - 05:14 As you can see, that's exactly what I 05:12 - 05:15 did. And now it does two things. This is 05:14 - 05:17 saved every time I want to use 05:15 - 05:18 perplexity later. And if I'm making a 05:17 - 05:20 recording like this, it's not going to 05:18 - 05:21 expose my API key. Especially if you're 05:20 - 05:23 like downloading your template and 05:21 - 05:24 sending it to someone else, you won't 05:23 - 05:26 have your API key actually there because 05:24 - 05:27 this will just reset. And as you can 05:26 - 05:30 see, I did the exact same thing with 05:27 - 05:31 OpenAI right here and with Tavi. And by 05:30 - 05:33 the way, because I'm not exposing my 05:31 - 05:34 credentials here, this entire workflow, 05:33 - 05:36 if you want to access it, will be 05:34 - 05:37 available for download in my free school 05:36 - 05:38 community. The link for that is down in 05:37 - 05:40 the description if you want to play 05:38 - 05:41 around with some of these hacks. And if 05:40 - 05:42 you're wondering how I did this, I hit 05:41 - 05:44 one to zoom out and I hit zero to zoom 05:42 - 05:46 back in. That was a little bonus hack 05:44 - 05:47 for you guys. Anyways, number 12 is 05:46 - 05:49 error workflows. You can use this native 05:47 - 05:50 integration error trigger and you can 05:49 - 05:52 link it up to any of your active 05:50 - 05:54 workflows and this trigger will 05:52 - 05:56 basically go off whenever that active 05:54 - 05:57 workflow errors. This way you can send 05:56 - 05:59 yourself email notifications, text 05:57 - 06:01 notifications, whatever you want if one 05:59 - 06:02 of your main workflow errors. And I 06:01 - 06:03 actually did a full video on this one 06:02 - 06:05 which you can check out right here. 06:03 - 06:06 Okay, number 13 is workflow variables. 06:05 - 06:08 And what do I mean by that? Let me show 06:06 - 06:11 you. You can access stuff like execution 06:08 - 06:12 ID, workflow name, workflow ID, workflow 06:11 - 06:14 active, yes or no. You can access this 06:12 - 06:15 kind of stuff with the workflow 06:14 - 06:16 variables. So, if I just grab this 06:15 - 06:18 trigger, hook it up, and hit test 06:16 - 06:19 workflow. When we come back in here, we 06:18 - 06:21 can see there's a little section right 06:19 - 06:23 here that says variables and context. 06:21 - 06:25 And it has stuff like dollar sign now, 06:23 - 06:26 dollar sign today, execution variables, 06:25 - 06:29 and workflow variables. All right, 06:26 - 06:30 number 14 is you can specify inputs when 06:29 - 06:32 you're sending data from one workflow to 06:30 - 06:33 a subworkflow. So, what you'll do is 06:32 - 06:35 you'll click into that actual trigger. 06:33 - 06:36 You can see that what you typically 06:35 - 06:38 would do is maybe just click on accept 06:36 - 06:40 all data, and this would put all of the 06:38 - 06:41 incoming data into a field called query. 06:40 - 06:43 But what you can do is you can come in 06:41 - 06:44 here and say, I want to define using 06:43 - 06:46 fields below. And then I can say, you 06:44 - 06:48 know, I want this workflow to receive, 06:46 - 06:51 you know, phone number, I want it to 06:48 - 06:53 receive email, and I want it to receive 06:51 - 06:54 uh customer name. And that's it. So, 06:53 - 06:56 it's going to be already put into those 06:54 - 06:58 three fields for you to map later, which 06:56 - 06:59 makes it a lot easier. And of course, if 06:58 - 07:01 you wanted to use some mock data, you 06:59 - 07:02 could hit this pencil and set those 07:01 - 07:03 three fields so that you can map the 07:02 - 07:06 rest of the workflow. All right, number 07:03 - 07:07 15, 16, and 17 all sort of go handinand. 07:06 - 07:09 So, let's take a quick look. What we 07:07 - 07:10 have in here is this AI agent is going 07:09 - 07:12 to send an email to 07:10 - 07:14 nate@acample.com asking if he enjoys 07:12 - 07:16 playing with sand. So I'm going to real 07:14 - 07:18 quick grab this manual trigger so we can 07:16 - 07:19 actually fire this off. Hook it up there 07:18 - 07:21 and hit test workflow. Okay. So what 07:19 - 07:23 this is going to do is it's going to use 07:21 - 07:25 this email agent which is a sub workflow 07:23 - 07:27 and it's going to say okay we sent that 07:25 - 07:28 message for you. So number 15 is looking 07:27 - 07:29 at subexecution data. So what we would 07:28 - 07:31 do is we'd click on this tool that 07:29 - 07:33 called and then right here we click on 07:31 - 07:35 view subexecution. So, this pulls up 07:33 - 07:37 that actual execution of the workflow 07:35 - 07:38 that we just called. And this is how we 07:37 - 07:39 can see if it actually did what we 07:38 - 07:40 wanted or not. And what you may notice 07:39 - 07:42 is that it didn't actually call that 07:40 - 07:43 send email tool. And we only would have 07:42 - 07:45 known that because we looked at the 07:43 - 07:47 subexecution data. Because if we come 07:45 - 07:49 back into the agent itself, it tells us 07:47 - 07:50 I've sent an email to Nate asking if he 07:49 - 07:52 enjoys playing with the sand. So, we 07:50 - 07:54 would now think that it's been done when 07:52 - 07:56 in reality it has not been done. So, 07:54 - 07:57 looking at hack number 16 is agent logs. 07:56 - 07:59 We would come into the agent logs. We'd 07:57 - 08:00 basically see what it did. First of all, 07:59 - 08:02 it used its brain to read through the 08:00 - 08:03 system prompt and then take action. 08:02 - 08:04 There's no system prompt for this agent 08:03 - 08:06 as you can see. So there wasn't much to 08:04 - 08:08 read. Then we can see it called the 08:06 - 08:10 email agent and then we can see exactly 08:08 - 08:12 what it sent over to that tool and what 08:10 - 08:13 it got back. And so because it sent over 08:12 - 08:15 a full email and not really any 08:13 - 08:17 directions, that's why that sub workflow 08:15 - 08:18 was kind of confused. And then it sent 08:17 - 08:20 us back this response which was the 08:18 - 08:22 email. And then the main brain gets back 08:20 - 08:23 that response from the tool and that's 08:22 - 08:24 why it thinks, okay, we sent that off. 08:23 - 08:26 So I'm going to let the human know that 08:24 - 08:28 we did. And then finally, when it comes 08:26 - 08:30 to viewing the subexecution data, you 08:28 - 08:31 can copy stuff to the editor. So, if I 08:30 - 08:34 go back into that subworkflow execution 08:31 - 08:35 right here, what I can do is click copy 08:34 - 08:37 editor, and this is going to pull in the 08:35 - 08:39 live data that we just saw in that 08:37 - 08:40 execution, and now we're able to play 08:39 - 08:41 with it. So, you can see this is the 08:40 - 08:43 query that was sent over from our main 08:41 - 08:45 agent, and we'd be able to remap stuff 08:43 - 08:46 in here and test it out to make sure 08:45 - 08:47 we're good. All right, number 18, we 08:46 - 08:48 have human in the loop node. So, if I 08:47 - 08:50 can just reactivate these real quick, 08:48 - 08:52 and if I hit tab, you can see right here 08:50 - 08:54 we have human in the loop nodes, which 08:52 - 08:55 basically says wait for approval or 08:54 - 08:57 human input before continuing. And you 08:55 - 08:59 can do this with Discord, email, Google 08:57 - 09:00 chat, all these things, all these 08:59 - 09:02 integrations. And basically what that 09:00 - 09:03 means is we have this operation where we 09:02 - 09:05 send off something in something like 09:03 - 09:06 Discord and we wait for the human to 09:05 - 09:08 respond. So let's take a look at what 09:06 - 09:10 that looks like in Slack real quick. In 09:08 - 09:11 this case, we can send just an approval, 09:10 - 09:13 which would just be a button that says 09:11 - 09:14 yes. We could change the type of 09:13 - 09:16 approval to have an approval and a 09:14 - 09:18 disapprove button rather than just an 09:16 - 09:19 approve button. We could also do a 09:18 - 09:21 custom form or free text, and that way 09:19 - 09:23 we're able to actually get feedback from 09:21 - 09:24 a human that's textbased rather than 09:23 - 09:26 just a yes or no button. I made a full 09:24 - 09:27 video breaking down this operation in 09:26 - 09:29 Naden. So if you want to watch that, 09:27 - 09:30 check it out right here. All right, 09:29 - 09:32 number 19 is we have the ability to 09:30 - 09:33 standardize inputs. So let me just 09:32 - 09:35 activate these two nodes real quick and 09:33 - 09:37 let me explain what's going on. So we 09:35 - 09:39 have an AI agent that's listening for a 09:37 - 09:41 field right here called 09:39 - 09:42 JSON.ext. And if we were to hit test 09:41 - 09:44 workflow and now our Telegram trigger is 09:42 - 09:46 listening for us and I pop this up and I 09:44 - 09:48 say hi, we can see that the agent's 09:46 - 09:50 going to get that message and it's able 09:48 - 09:53 to process it because it comes through 09:50 - 09:54 Telegram in a field called JSON.ext. But 09:53 - 09:56 if we were to open up the native end 09:54 - 09:58 chat and say hi, this is going to error 09:56 - 10:00 out because there's no prompt specified 09:58 - 10:02 because what comes through in the native 10:00 - 10:04 chat is a field called chat input. So 10:02 - 10:06 the idea here is to standardize the 10:04 - 10:07 input so you can have multiple forms of 10:06 - 10:09 communication with an agent. So all I 10:07 - 10:12 have to do is hook up this set node that 10:09 - 10:13 I built, hook that into our NN AI agent, 10:12 - 10:15 and what that's going to do is take this 10:13 - 10:17 message and put it in a field that 10:15 - 10:19 matches this field and we'll be good to 10:17 - 10:21 go. So if I say hi now we can see the 10:19 - 10:22 agent's going to be able to process it 10:21 - 10:23 because it comes through in a field 10:22 - 10:26 called JSON 10:23 - 10:27 message.ext. Okay. Number 20 is the 10:26 - 10:30 option to continue on error. So what we 10:27 - 10:32 have right here is three URLs within 10:30 - 10:34 this um code node. McDonald's Chipotle 10:32 - 10:36 and a fake URL. So if we hit test 10:34 - 10:38 workflow, we can see that it's going to 10:36 - 10:39 loop through each of those and it's 10:38 - 10:42 going to finish out and basically say 10:39 - 10:44 okay here is the, you know, HTML content 10:42 - 10:46 from that first one. Here's the third 10:44 - 10:47 one and then here's that last one. But 10:46 - 10:50 what you didn't see is that that third 10:47 - 10:52 run of the false URL didn't actually 10:50 - 10:54 work. We got a 404 error. But the first 10:52 - 10:55 two are coming back with actual HTML. 10:54 - 10:57 And so the reason this didn't stop the 10:55 - 10:58 whole workflow is because in this node 10:57 - 11:00 in the settings, we're telling it to 10:58 - 11:02 continue on error rather than stopping 11:00 - 11:03 the entire workflow. So if I change it 11:02 - 11:04 back to stop workflow, what it's going 11:03 - 11:06 to do is it's just going to break and 11:04 - 11:08 then the automation is never going to 11:06 - 11:10 finish. As you can see, even though the 11:08 - 11:11 third one didn't work, but the first two 11:10 - 11:12 still did. So, imagine you had a lead 11:11 - 11:14 list of a thousand people that you were 11:12 - 11:16 trying to process and the a thousandth 11:14 - 11:18 one failed. You would lose pretty much 11:16 - 11:19 all that data. Well, you really wouldn't 11:18 - 11:22 because, as we know, you can copy it to 11:19 - 11:23 the editor and still access it, but it 11:22 - 11:25 would be annoying. And then just to stem 11:23 - 11:28 right off of that, you can also have it 11:25 - 11:29 continue on error with an error output. 11:28 - 11:31 So, that's just another option in the 11:29 - 11:32 settings, as you can see, using error 11:31 - 11:34 output. So, that would basically mean 11:32 - 11:35 anything that errored would go off down 11:34 - 11:37 this other branch. So, we could 11:35 - 11:39 basically say, okay, anything that's 11:37 - 11:40 good, we'll send back this way. But if 11:39 - 11:41 it errors, we're just going to send 11:40 - 11:43 ourselves an email and say, "Hey, this 11:41 - 11:45 URL errored." And it doesn't have to be 11:43 - 11:47 just like an HTTP request or an AI agent 11:45 - 11:49 node. Pretty much every node has this 11:47 - 11:50 functionality within naden. Okay, so 11:49 - 11:52 moving on to some other settings that 11:50 - 11:53 you can play around with in these nodes. 11:52 - 11:54 They're all a little bit different based 11:53 - 11:55 on the node you're playing with. So 11:54 - 11:57 definitely just take a look when you're 11:55 - 11:59 setting up these workflows. But in your 11:57 - 12:00 settings, you can have SSL certificates 11:59 - 12:02 if it's with an HTTP request. You can 12:00 - 12:04 have it always output data because 12:02 - 12:05 sometimes it might just like not return 12:04 - 12:07 anything, but you still want the process 12:05 - 12:08 to continue on. You can have it only 12:07 - 12:10 execute once. And I've actually had some 12:08 - 12:11 workflows where I didn't really know 12:10 - 12:13 this was there. So I would just have 12:11 - 12:14 three items coming through. I'd limit it 12:13 - 12:16 to one and then I'd continue on. But all 12:14 - 12:18 I'd have to do is click on execute once. 12:16 - 12:19 You can also have stuff retry and fail 12:18 - 12:21 and you get to specify, you know, like 12:19 - 12:23 what to do, how many times, how much you 12:21 - 12:24 wait. But like I said, each node has 12:23 - 12:26 kind of different settings. So in here 12:24 - 12:28 we have, you know, we don't have the SSL 12:26 - 12:29 certificates one obviously. In here we 12:28 - 12:31 have one called duplicate item. So you 12:29 - 12:32 can basically, you know, tell it how 12:31 - 12:34 many times to duplicate stuff. And I 12:32 - 12:35 think this is primarily used for testing 12:34 - 12:37 and debugging. And then of course you 12:35 - 12:38 can leave notes down here which don't 12:37 - 12:40 actually change anything about the 12:38 - 12:41 functionality of the node. It's just in 12:40 - 12:43 case you're you know sending over a 12:41 - 12:44 template or someone on your team has to 12:43 - 12:46 look at this workflow, you can leave 12:44 - 12:47 some important notes right there. Okay. 12:46 - 12:49 Number 23 is that we can have structured 12:47 - 12:51 outputs with our agents. So right here 12:49 - 12:52 I'm telling this guy to output four 12:51 - 12:54 stories. And we're going to see it 12:52 - 12:56 outputs four different scenes all in one 12:54 - 12:57 item, all in one object. So as you can 12:56 - 12:59 see right here, these are our four 12:57 - 13:01 different stories. The Missing Key, The 12:59 - 13:03 Last Leaf, Midnight Bakery, Quantum 13:01 - 13:06 Leap. What I can do is right here I turn 13:03 - 13:07 this switch on which says require 13:06 - 13:09 specific output and it creates this new 13:07 - 13:11 little tool for us and I drag that into 13:09 - 13:13 my structured output parser which 13:11 - 13:15 basically says output stuff like this 13:13 - 13:17 and you can have chachebt generate the 13:15 - 13:19 JSON example that you need and now if I 13:17 - 13:21 run this guy it's going to output it in 13:19 - 13:22 four different little objects. So there 13:21 - 13:24 we go just finished up and now we can 13:22 - 13:25 see we have four scenes and now I can 13:24 - 13:27 sort of like split those out and process 13:25 - 13:29 them individually or whatever I want to 13:27 - 13:31 do with them. Okay so number 24 is a 13:29 - 13:33 little function we can do with an array. 13:31 - 13:35 So in here we have a list of foods which 13:33 - 13:36 is apple banana steak. And then if I 13:35 - 13:37 come into here, if we're basically 13:36 - 13:39 outputting that, it comes through in a 13:37 - 13:41 weird ugly way called apple banana 13:39 - 13:43 steak. So if I open this up, what I can 13:41 - 13:45 do with an array is at the end I do dot 13:43 - 13:46 join. And now you can see it's going to 13:45 - 13:48 separate them by commas. But what we can 13:46 - 13:50 do is within this little parenthesis, we 13:48 - 13:52 can tell it how do we want this array to 13:50 - 13:53 be separated. So I just have to put two 13:52 - 13:54 quotes and all of a sudden they're 13:53 - 13:56 they've been combined. So I could put a 13:54 - 13:58 space, I could put two spaces, I could 13:56 - 13:59 say and and then I could put more 13:58 - 14:01 spaces. And you can see that's how we 13:59 - 14:03 can split up these items. You can also 14:01 - 14:05 do a slash and an N to create a new 14:03 - 14:06 line. And you can also do multiple. So 14:05 - 14:07 you can have them like really spaced out 14:06 - 14:09 if you want. As you can see. All right. 14:07 - 14:11 And then last but not least, we have the 14:09 - 14:13 ability to replace things. So the most 14:11 - 14:15 common example I use is replacing double 14:13 - 14:16 quotes with something else. So right 14:15 - 14:18 here we can see we have the query 14:16 - 14:20 pineapples on pizza. And we're trying to 14:18 - 14:21 search the web for pineapples on pizza. 14:20 - 14:23 But as you may know, it's not going to 14:21 - 14:25 work to have, you know, basically double 14:23 - 14:27 quotes inside double quotes in a JSON 14:25 - 14:28 body. So, all I'd want to do is I'm 14:27 - 14:30 going to go in here and I'm going to 14:28 - 14:33 grab this function real quick. Go back 14:30 - 14:35 into the HTTP request, open up our 14:33 - 14:37 expression, and paste that at the end. 14:35 - 14:39 And that's basically just removing those 14:37 - 14:40 double quotes as you can see. So, now if 14:39 - 14:42 I run this, it would actually be able to 14:40 - 14:43 search the web for pineapples on pizza. 14:42 - 14:45 And there you go. You can see we got our 14:43 - 14:47 answer back from Tavi. And there you 14:45 - 14:49 have it. Those were 25 Nin hacks that 14:47 - 14:50 are hopefully going to increase your 14:49 - 14:51 productivity. If you enjoyed this one or 14:50 - 14:53 you learned something new, leave a like. 14:51 - 14:54 It definitely helps me out a ton. And as 14:53 - 14:57 always, I appreciate you guys sticking 14:54 - 14:58 around to the end of this one for hack 14:57 - 15:01 number 25. I'll see you guys in the next 14:58 - 15:01 one.

Summary

這部影片介紹了 25 個 Naden 的實用技巧，旨在提升開發者的生產力。這些技巧從基礎概念到進階應用，涵蓋了資料處理、工作流程管理、除錯以及與外部服務的整合等多個方面。影片強調了這些技巧能幫助使用者更有效率地建構和維護 Naden 工作流程。

Notes

  • 固定值與表達式轉換： 在輸入框中，在輸入任何內容前，直接按下鍵盤上的等號 (=) 即可將固定值切換為表達式。
  • 固定 (Pin) 資料： 對於會產生費用或執行時間長的節點，可以將其輸出資料固定，避免重複執行 API 調用，節省成本和時間。
  • 模擬資料 (Mock Data)： 在節點或觸發器中，可以手動填入模擬資料，方便測試工作流程的後續部分，尤其適用於與其他工作流程觸發的場景。
  • 節點釘選快速鍵： 選中節點後，按下鍵盤上的 'P' 鍵可以快速釘選或取消釘選資料。
  • 節點間快速切換： 在節點設定面板中，將滑鼠移到節點的左右邊緣，即可直接切換到相鄰節點的設定，無需退出。
  • 使用方向鍵和 Enter 鍵導航節點： 在工作流程視圖中，可以使用方向鍵在節點間移動，並按下 Enter 鍵打開選中的節點。
  • 停用 (Deactivate) 節點： 可以透過點擊節點的選單或選中節點後按下 'D' 鍵來停用節點，這在測試工作流程時，用於排除錯誤很有幫助。
  • 組織化技巧：
    • 便利貼 (Sticky Notes)： 使用便利貼來標註工作流程的說明或指南，支援 Markdown 格式，並可自訂顏色以區分不同類型的節點。
    • 命名節點： 快速命名節點，可以透過點擊節點並按下 F2 鍵來達成。
  • 工作流程歷史記錄： 可以查看和還原工作流程的歷史版本，類似於 Google 文件。
  • 快速新增節點： 在工作流程中，按下 Tab 鍵即可快速調出新增節點的選項。
  • 禁用歸因 (Attribution) 訊息： 在發送電子郵件、Telegram 或 Slack 訊息等整合節點時，可以禁用「訊息由 Naden 發送」的歸因文字。
  • 獲取當前日期和時間： 在表達式中，使用 {{ $now }} 函數可以獲取當前

從混亂到清晰

拖動滑桿查看差異。停止觀看，開始理解。

LKK 的方式

一場 25 分鐘的豪賭

費力尋找答案
快轉廣告和廢話
被標題黨分散注意力

PeekaTube 的方式

7 秒鐘總結

立即獲取重點
直達相關時間戳
實現零干擾學習

從此告別標題黨！

Peekatube 為任何帶有 CC 字幕的 YouTube 影片提供即時 AI 摘要，讓您在點擊前就掌握影片精華。

免費測試版

免費測試版用戶

搶先體驗測試期間的所有功能

$0測試期間
  • 帶有 CC 字幕的 YouTube 影片快速摘要
  • 測試期間無限制使用
  • 測試版功能：
  • 搶先體驗新功能
  • 一鍵匯出摘要與逐字稿
  • 優先客服支援
  • Chrome 擴充功能（即將推出）
  • 正式上線時享有優惠定價

常見問題

關於 PeekaTube 的一切，這裡都有答案

PeekaTube 如何運作？
很簡單。您提供 YouTube 連結後，AI 會讀取影片的字幕 (CC) 或逐字稿，分析內容的關鍵論點，並生成一份架構清晰的摘要。
只能用在有 CC 字幕的影片上嗎？
目前是的，摘要的品質高度依賴影片字幕的準確性。不過，支援「無字幕影片」的技術已在開發中，如果您非常需要此功能，可以
什麼類型的影片摘要效果最好？
PeekaTube 特別擅長處理以「談話」為核心的內容，例如：教育演講、技術教學、人物訪談、紀錄片和新聞報導。基本上，只要影片有字幕，AI 就能進行分析。
摘要的準確度如何？
我們旨在高準確度地捕捉影片的核心思想和架構。然而，請將它視為一個聰明的「觀看指南」，而非完全取代觀看。它非常適合用來判斷一部影片是否值得您投入時間，並快速掌握其核心概念。
支援哪些語言？
PeekaTube 不僅能處理多種來源語言，更能將摘要翻譯成您指定的語言。例如，您可以為一部日文影片生成中文摘要。目前我們支援的輸出語言包含：中文、英文、日文、韓文、西班牙文、法文、德文、義大利文、葡萄牙文等。
處理速度可以更快嗎？
我一直在努力優化速度！目前的處理時間，包含了確保服務穩定性的必要步驟。為了防止服務被 YouTube 封鎖 IP，後端使用了代理 (Proxy) 來確保 PeekaTube 能為您持續穩定運作。這個做法雖然會增加幾秒鐘的讀取時間，卻是維持服務不中斷的關鍵權衡。我的目標，是在不犧牲穩定性的前提下，持續提升處理速度。
這個工具適合誰使用？
PeekaTube 適合任何珍惜時間的人： - 學生：用來做報告研究、複習課程 - 專業人士：快速跟進產業趨勢和教學影片 - 終身學習者：探索新知，無需冒著浪費時間的風險
未來的開發計畫是什麼？
我的首要任務是推出大家期待已久的「Chrome 擴充功能」，讓 Pro 用戶能直接在 YouTube 頁面上使用。此外，如前所述，支援「無字幕影片」的技術也在積極開發中。您的支持，是讓這些功能提早實現的最大動力！

還有其他問題？

我們很樂意為您解答任何疑問