# Claude Code + NotebookLM + Obsidian: The Research Stack Nobody's Using  **Channel:** Artem Zhutov **URL:** https://www.youtube.com/watch?v=qiOu7Ptjxng&t=527s **Date processed:** 2026-02-24 --- ## Summary This video demonstrates a powerful research workflow that integrates Claude Code, NotebookLM, and Obsidian. The core idea is to overcome Claude's context limitations by using NotebookLM as a research intermediary—Claude Code controls NotebookLM via CLI, uploads sources (YouTube videos, PDFs, articles, vault notes), asks questions, and imports the grounded, cited answers back into Obsidian. This creates a closed loop where sources live in NotebookLM, answers with citations flow back to Obsidian, and everything is connected via wiki-links and knowledge graphs. The workflow addresses several pain points: Claude can't read 300 files at once, manually managing research tabs is unscalable, and LLM hallucination undermines research reliability. By routing queries through NotebookLM's citation system and importing results into Obsidian, users get grounded answers with clickable citations that trace back to exact passages in source material. The video shows practical use cases including academic research (arXiv paper discovery), company knowledge bases, podcast generation, flashcard creation, and market research. The system is built on a custom Claude Code skill and NotebookLM CLI tool, both available via links in the video description. The presenter also runs workshops teaching users how to set up this workflow in their own vaults. --- ## Key Points - **Core workflow:** Claude Code → NotebookLM CLI → NotebookLM (process sources) → Claude Code (import cited answers) → Obsidian (knowledge graph) - **Context limitation workaround:** Claude can't read 300 files, but NotebookLM can—so use NotebookLM as a grounded research layer - **Citation grounding:** NotebookLM provides exact citations for every claim, reducing hallucination risk and making research verifiable - **Programmatic source curation:** Claude Code searches YouTube, web, vault notes, and adds them to NotebookLM notebooks—no manual tab management - **Obsidian integration:** Answers are imported with wiki-link citations pointing to exact passages in source notes, creating navigable knowledge graphs - **Multiple output formats:** Audio overviews (podcasts), flashcards, mind maps (Canvas), slide decks—all generated from NotebookLM and imported to Obsidian - **Use cases:** Academic research (arXiv searches), company onboarding knowledge bases, market research, podcast analysis, daily note analysis - **Dashboard-centric workflow:** Create research dashboards in Obsidian showing sources, Q&A logs, citation frequency, topic connections --- ## Detailed Notes ### Introduction [0:00 - 1:21] - **Problem statement:** Claude's context window can't handle 300 files at once - **Solution:** Use NotebookLM as intermediary—reads everything, cites sources, controlled by Claude Code from terminal - **Demo preview:** 20 sources processed, emerging themes query, cited answers imported to Obsidian with clickable citations to exact video passages ### What is NotebookLM? [1:39 - 2:34] - Google research tool for uploading sources and creating notebooks - Supports multiple source types: videos, articles, PDFs, etc. - Can generate artifacts: audio overviews, mind maps, video overviews - **Key feature:** Exact citations from sources, reducing LLM hallucination - Makes Claude Code research "more reliable" by grounding answers in verifiable sources ### Integration Architecture [2:34 - 4:00] - **Claude Code ↔ NotebookLM CLI ↔ NotebookLM** - GitHub project enables CLI interaction (link in video description) - Claude Code can: - Upload sources (YouTube links, URLs, blog posts, Obsidian notes) - Ask questions via terminal - Get cited responses - Export answers back to Obsidian - **Advantage over web UI:** Responses saved in Obsidian, knowledge graph visualization, no manual copy-paste ### Live Research Demo [4:00 - 7:18] - **Research topic:** Claude and Obsidian integration - **Workflow:** 1. Use YouTube search skill to gather video links 2. Claude Code creates new NotebookLM notebook 3. Adds 20 curated videos as sources (added in parallel) 4. Asks question about "gaps in the space" 5. Creates Obsidian dashboard to track sources - **Source curation advantage:** More granular control than web search, higher signal-to-noise ratio - Dashboard shows: all sources, summaries, Q&A log, citation tracking ### Citations and Grounding [7:52 - 11:15] - Response imported to Obsidian with exact citations - **Example:** "Claude can automatically add YAML front matter" → links to specific video passage - Wiki-links point to timestamped sections of source transcripts - **Citation accuracy audit:** ~60% strong match, 31% partial match, 10-15% weak match (presenter considers this "quite strong grounding") - Citations displayed in graph view showing connections between sources and topics ### Obsidian Bases Integration [12:00 - 13:05] - Each NotebookLM source becomes a file in Obsidian - Can create Obsidian Base (database view) for sources - Fields: topics (extracted by NotebookLM), type (YouTube), status, citation references - Topics auto-generated by NotebookLM, displayed as emerging themes ### Graph View Exploration [13:08 - 14:06] - Filter by research dashboard note to see all connected sources - Topics become hub nodes (e.g., "Claude Code" node connects to 6 videos mentioning it) - Enables discovering gaps and new ideas by exploring topic connections ### Artem's Lab Program [14:07 - 15:37] - 6-week program: Obsidian + Claude Code integration - Weekly workflow builds from scratch - Addresses: context management, thought capture, reducing repetitive explanations to Claude - Workshop + mastermind format - Also running: "Zero to Obsidian CLI Pro in 90 Minutes" workshop ### Adding Multiple Source Types [15:40 - 18:15] - **Beyond YouTube videos:** Can add web articles, PDFs, vault notes - Demo: Claude Code searches web for articles on "Claude Code and Obsidian," adds 3 articles + Obsidian vault file - Sources appear in real-time in NotebookLM notebook - Example: Added claude.md setup guide and course outline from vault ### Scaling to 300 Sources [18:36 - 19:11] - NotebookLM supports up to 300 sources - Demo: Uploaded 300 daily notes, can chat with them to extract patterns - **Use case:** Personal knowledge base querying ### Persona Configuration [19:11 - 20:01] - Can configure NotebookLM persona for response style - Demo: "Deliver concise responses, no filler, no preamble" - Tested with 3 parallel questions - Result: Much shorter, direct answers ### Importing Q&A with Citations [20:01 - 21:03] - Use `--json` flag to get citation resolution - Citations extracted and resolved automatically - Answers appear in Obsidian Q&A log with wiki-link citations - **Example citation check:** Question about "Claude context management" correctly pulled from Teresa Taurus video with exact advice quote ### Citation Mechanics [21:03 - 22:14] - NotebookLM divides sources into chunks with IDs - Citations reference chunk IDs (e.g., [39]) - Wiki-links map to headers in source notes → click to jump to exact passage - **Old workflow (pre-integration):** Open browser, 20 tabs, manual drag/drop, lose citations, "just forget about it" - **New workflow:** Programmatic source addition, curated lists, Claude can dig deeper, full citation traceability ### Audio Overview (Podcast Generation) [22:55 - 25:00] - NotebookLM can generate audio overviews (podcasts) from sources - Claude Code command: "Generate audio overview focused on the gaps, add to Obsidian vault" - MP3 file embedded in notebook note - **Use case:** Listen on phone via Obsidian Sync while mobile - Custom podcasts generated on demand from research topics ### Other Output Formats [25:01 - 25:56] - **Mind maps:** Generated from NotebookLM, exported to Obsidian Canvas - **Flashcards:** 58 cards example, clickable Q&A format for studying - **Slide decks:** Possible but presenter "not a big fan" (can't edit easily) ### Academic Use Case [26:28 - 27:30] - **arXiv integration:** Claude Code can search arXiv for papers on specific topics (e.g., "quantum computing with Rydberg atoms") - Feeds papers into NotebookLM for grounded citations - **Critical for academics:** Responses not made up, full observability over sources - Can specify date ranges for foundational vs. recent research ### Additional Use Cases [27:30 - 29:03] - **Market research:** Competitor blog posts, YouTube videos - **Company knowledge base:** Onboarding docs in NotebookLM → new employees ask basic questions without consuming team attention - **Shared wiki:** Company-wide knowledge repository - **Podcast analysis:** Lex Fridman 4-hour episodes → research AI field trends ### Closing the Loop [29:03 - 30:17] - **Old problem:** NotebookLM outputs trapped in browser, copy-paste to scattered systems (Google Keep, Obsidian), no consistency - **New solution:** Consistent research dashboards in Obsidian, all data in markdown files on local computer - Claude Code can re-engage with saved sources anytime - Full loop: Notes → NotebookLM → Knowledge graphs → Obsidian ### Dashboard-Centric Workflow [30:17 - 31:47] - Dashboard note = centralized view of research topic - Contains: sources list, Q&A log, citation frequency - **Obsidian CLI integration:** Claude Code can read dashboard via Obsidian CLI, extract Q&A, present as table - Note: Bases don't work with CLI yet, need to translate to Dataview ### Graph Visualization Finale [31:47 - 32:42] - Source files have wiki-link topics (e.g., #claudecode, #secondbrain) - Graph view: Add group "NotebookLM source" (colored red) to distinguish sources from topics - Visual knowledge map: Red = sources (videos, articles), other colors = topics - Can explore topic connections across source types ### Conclusion [32:42 - 33:46] - **Tools:** NotebookLM Python library + Claude Code skill (links in description) - **Workshop:** Wednesday session—build workflow in your vault - **Obsidian CLI + Claude Code workshop:** Free program available - Releases one workflow video per week --- ## Connections *Links to related vault notes will be added here* --- ## My Notes *Space for your own thoughts, reactions, and connections after watching* --- ## Full Transcript <details> <summary>Click to expand transcript</summary> [0:00] Clo can't read 300 files at once. The [0:03] context you know is just not big enough. [0:05] So I put it inside notebook where it [0:07] reads everything sites every source and [0:10] cloud code controls it from a terminal. [0:12] Let me show you how it works. Here is [0:14] the example of research which I did with [0:18] notebook lamb where you had 20 sources [0:21] and here is a question which I asked [0:23] about emerging themes across those 20 [0:26] sources and here is actual response from [0:29] notebook lm imported to our obsidian [0:31] vault where we have um exact citations [0:35] from the videos. Here is example of the [0:39] citation. I can click on that and it [0:42] guides me to exact passage from the [0:46] video transcript. Now our answers are [0:49] grounded and here are the all the [0:51] citations from this video now. And here [0:55] I have a sources referenced by the [0:57] frequency of the citations. And here is [1:01] a quick diagram to explain what's going [1:03] on. We had a cloud code which connects [1:06] to nodem. it it uploads those uh 20 [1:09] videos there and then you can ask the [1:12] questions about it literally anything [1:14] and then you get a response from [1:16] notebookm with the cited answers and we [1:19] import those into our obsidian. Now in [1:21] this video what we're going to go [1:22] through is um a research workflow from [1:26] cloud code integrated with obsidian how [1:29] to get sources and citations coming to [1:32] Obsidian. how to do audio overview and [1:35] podcast generation from the cloud code [1:37] and other use cases. So let's get [1:39] started. First, what is notebook? Uh [1:43] notebookm is a Google uh research tool [1:45] where you can upload a bunch of sources [1:48] and you have those notebooks with [1:50] sources and here is example. You have [1:54] all of the uh sources which you can [1:57] upload. You can add them, you can chat [2:00] with them, and you can generate [2:01] artifacts such as audio overview, uh, [2:05] mind maps, video overview. And the key [2:08] here is that we get those exact [2:12] citations from the videos or from any [2:15] sources which we are uploading. So our [2:19] answers are grounded. And this partially [2:22] solves this issue where LLMs hallucinate [2:26] the answers and making the uh research [2:30] with uh cloud code more reliable. Now [2:34] here I integrated cloud code into this [2:36] picture and cloud code can interact with [2:39] nuclem through a common line interface [2:43] and here is a GitHub project which [2:47] enables this uh interaction of the clot [2:50] code. We have a various commands which [2:53] cloud code can do based on our um [2:56] notebook lm notebooks and and and we [3:01] going to go through them. Uh what what's [3:03] possible here is a diagram that cloud [3:06] code can call this notebook LM CLI to [3:10] interact with um actual notebook LM to [3:13] ask questions to get the responses. It [3:16] can also upload the sources such as [3:18] YouTube links, URLs, blog posts, all [3:20] like even your obsidian nodes. And the [3:23] advantage here is of using uh it with a [3:27] cloud code through CLI is that you can [3:32] actually integrate it with your [3:34] Obsidian. You can actually get those [3:37] responses back to your Obsidian and work [3:40] there. So it becomes much much more [3:43] convenient rather than using their uh [3:45] web UI. You can actually save responses, [3:48] you can ask questions and you can um use [3:52] those knowledge graph uh to display um [3:57] connections between different sources. [4:00] All , so let's try something. We're [4:03] going to go and try to do a research [4:06] topic. I'm starting a cloud code inside [4:09] of my Obsidian world and I have this [4:11] notebook LM skill which enables this [4:14] programmatic interaction. So CL can [4:16] actually learn how to use uh this tool [4:19] which we introduced. I'll just execute [4:22] this skill and um so let's say I want to [4:25] research about um uh clo and obsidian [4:29] integration and I want to use my YouTube [4:33] skill to gather the links for the [4:36] YouTube videos. And then I want to pick [4:39] let's say like 10 videos to process. And [4:43] then we're going to create uh the [4:45] sources. And then we can try to ask what [4:51] are the gaps in those videos. What are [4:53] the opportunities for um exploring [4:56] different ideas and the workflow here is [5:00] this custom skill called YouTube search. [5:03] It loads the skill and now it searches [5:06] uh different videos uh from the YouTube. [5:10] Here are all of the responses, all of [5:13] the videos. We just search a bunch of [5:16] these. Uh what's very very cool and then [5:20] and then actually we're going to create [5:22] a new notebook. So once we are ready, [5:25] let's actually go back to notebook and [5:28] confirm with cloth. So what's going on? [5:31] Okay. Um, here are some unique relevant [5:34] videos. Okay, I already lo that. Cool. [5:38] Okay, here are the pool of the results. [5:41] Created list of the most relevant [5:42] videos. Nice. Nice. Sorted by relevance [5:47] and diversity perspective. We got 20 [5:50] videos. 20 videos. Okay. Um, that's [5:54] actually sounds very cool. Let's um try [5:57] to add those like create a new notebook [6:00] LM notebook and then add 20 of those [6:04] videos uh into there and ask the [6:07] question about emerging ideas or the [6:10] gaps which are exist in the space. Okay, [6:13] great. Um and here we have the skill [6:17] which um imports uh the sources. So it [6:21] teaches how to use this notebook lm [6:23] command line interface clo is going to [6:27] create a notebook right now and that's a [6:29] new notebook codex obsidian research. [6:32] Now let's see uh it adds the sources in [6:35] parallel. Okay we have added the source [6:38] that's very cool and you can see already [6:41] that they're appearing right here. I [6:43] don't know there are six videos. Nice. [6:46] There you go. More videos are coming. [6:50] That's cool. And I really love this [6:52] process because I can actually pick the [6:54] videos and it's much better than using [6:56] this f like web search or fast search [7:00] because I have like much more granular [7:01] control over the sources and this way [7:04] you can increase the signal to noise [7:06] ratio to a high high value rather rather [7:10] than relying on this generic web web [7:12] search for the adding a new sources. [7:15] Okay. So that's very cool. And um right [7:18] now Cloud Code creates for us a [7:20] dashboard. So we can actually look and [7:24] see all of this information in our [7:27] Obsidian vault. And that's the most [7:29] exciting part because I can actually [7:32] interact with all these sources within [7:34] my Obsidian. I can see them. I can see [7:37] connections between them and I can [7:39] actually ask questions here from a from [7:42] a terminal and cloud code will be able [7:44] to export those answers with cited [7:49] sources into our obsidian. Now let's [7:52] wait um let's wait a bit. It's importing [7:56] those sources and this skill is [7:58] available uh so you can just grab a link [8:01] in the description. Okay, we have [8:03] imported 20 sources and we want to [8:05] create a dashboard right now where we [8:07] can clearly see and track all of these [8:09] sources because it it's very very hard [8:13] um to navigate this and here is our [8:16] dashboard where we have all of these 20 [8:19] videos and we can see like their summary [8:23] uh from the no blue colm and right now [8:25] here we have Q&A log where we can log [8:28] the questions which you're asking and [8:31] we're going to log the response here as [8:33] well. Yeah, here we don't really need [8:35] this. All right, so it takes a bit of [8:38] time because um the model is thinking. [8:41] Um it's not that clock code is slow. [8:44] It's just um the Gemini model is um [8:47] takes some time to execute and actually [8:50] the response should appear here in the [8:52] chat after it runs. Oh, cool. I think [8:56] it's done. Okay, so we got the answer. [8:59] And um all right so here is the actual [9:03] um response from the notebook alm it's [9:06] in our obsidian vault and meanwhile I am [9:09] um asking clo to analyze if the [9:12] citations make sense but uh we can just [9:15] actually like go through them and see [9:18] what's going on. So here we are saying [9:21] that clo can automatically add YAML [9:24] front matter generate tax move files [9:26] into the folders and it references this [9:29] video. So let's actually see what's [9:32] going on under the hood here and we see [9:34] that uh okay we can mention it mentioned [9:38] that we have like this nicely formatted [9:40] files. Um good. Um now there is another [9:46] passage. Yes, you can also do the tags. [9:50] Okay, it kind of makes sense. We can [9:52] also see here tagging as well. Um, [9:55] overall it's it's quite good. Yes, the [9:58] sources make sense. Let's try to do [10:00] something else. Here is um to-do list [10:04] generator. Um, okay. The code can to [10:07] scan a wall to build a today to-do list [10:10] or daily briefing. Here is a video and [10:13] oh, okay. So that's um from the video [10:16] from tresa taurus that we are saying we [10:20] are using this today command and clo [10:22] goes going to trell and um getting the [10:27] information from there okay and [10:29] generates to-do list okay great that [10:32] makes sense now let's try to see about [10:35] something under exposed [10:38] uh AI agents can edit override or per [10:40] permanently delete your files uh here [10:43] we're mentioning the crowd operation or [10:46] the vault makes sense. Uh so then there [10:49] is what is this? G version control solid [10:53] AI edit every AIA distract. So let's see [10:57] what there um okay so here yes we have [11:01] mentioning of g that every action of [11:04] cloud is automatically logged. Great. So [11:07] overall no the references make sense. So [11:10] like our actual response is grounded [11:13] with the real data from the real videos. [11:15] Here is the response from clot. Oh, [11:18] nice, nice, nice, nice. Uh, I also run [11:21] this uh citation accuracy audit and here [11:24] is rating like strong match like around [11:28] 60%, partial match is 31% and weak match [11:33] is like let's say 10 to 15%. Now overall [11:37] it's quite quite good quite quite good [11:41] and it's quite strong grounding. And now [11:44] you can see this on a graph as well. [11:48] Here all your sources sited. You can go [11:50] to each of them. It has a file itself. [11:53] And you can actually integrate this with [11:56] obsidian bases because uh each um each [12:00] source is a file. We can ask lot I want [12:04] to build obsidian base for my uh [12:06] notebook sources. So I can display those [12:09] and add relevant fields for this view. [12:12] Can you also check how we do those [12:14] obsidian bases um and add this to our [12:18] dashboard also we extract topics uh from [12:22] from the notebook alm source and those [12:25] are generated by notebook alm and we can [12:27] actually see the emerging topics right [12:30] here. Um, okay. And meanwhile, Claude [12:33] have just created uh for us for us this [12:36] uh dashboard with all of our uh sources [12:39] and um nice. I wonder if you can add uh [12:44] more properties displayed in all sources [12:47] view and here are whole database of the [12:51] sources which we have. You have topics, [12:54] you have type YouTube, you have a status [12:57] for those and you can see where is this [13:01] uh source was cited in which question. [13:05] Now another opportunity is to look at [13:08] the graph view to see the connections [13:10] here. Here is my research file which I [13:15] put and I'm filtering by by this file [13:18] and I can see all of the different uh [13:20] connections across different sources. [13:22] Here is um my main dashboard [13:28] note and you can see it connects to to [13:31] this source about project corn kernels [13:33] and this project kernels is connected to [13:38] to other topics. Now where it gets [13:40] interesting when there are like multiple [13:42] uh like where there are similar topics [13:45] for example for this video we have a [13:47] cloth code and it connects to this uh [13:50] node called cloud code and you can [13:52] explore there what's going on and here [13:55] it connects to other six videos where we [13:59] mentioning cloud code. Now you can [14:01] explore this graph and bring insights [14:04] and new ideas and find the gaps. Now [14:07] this whole setup of uh of CLI skill and [14:11] wil weekly resolution I build all of [14:13] this in my lab. Uh and every week we [14:17] pick a workflow and build it from [14:19] scratch and I share it. And if you want [14:21] to set up it on your vault or build um [14:24] like something that tailors your needs, [14:26] here is a link uh you can go there and [14:29] here is a a six weeks program where we [14:32] explore integration of obsidian and clo [14:34] code to how clo can be applied to [14:37] different areas uh to helps you to help [14:40] you to be uh focused uh to actually um [14:45] stop explaining every time you go to uh [14:49] cloth Um what's the context? Um so you [14:53] can we address this issue in the lab. We [14:56] discuss how to capture thoughts and how [14:58] to have them organized. Um here we have [15:02] a workshop and mastermind. [15:04] And um we're going to learn about [15:07] Obsidian Clob Obsidian CLI and here you [15:11] can look at the schedule and also uh [15:15] reviews and people being engaged with [15:18] the material. and feel free to drop me a [15:21] question uh if you have any questions in [15:24] the WhatsApp. I'm also running this [15:26] workshop this week zero to obsidian CLI [15:30] pro in 90 minutes. If you're interested [15:32] in how to use Obsidian CLI and cloud [15:34] code together, feel free to join. Okay. [15:37] And let's take a step back about what [15:40] have just happened with all of these uh [15:43] citations and and and not essentially [15:46] what we did is okay so we got results [15:51] from YouTube we got five videos added to [15:54] our like actually like 20 videos added [15:57] to our notebook and then we asked um [16:01] from the terminal it's actually cloud [16:03] code asked a question for us about the [16:07] gaps [16:07] And we got response back with a [16:10] citations which makes sense into our [16:12] obsidian. Now there's no need to copy [16:15] paste no need to open like 20 tabs 20 [16:19] you know videos cloud can search it for [16:22] us and um add them to notebook [16:25] communicate and get results back to [16:27] oursidia. That's truly truly exciting. [16:30] Now and this scales uh beyond just [16:33] YouTube videos. You can think about your [16:35] vault notes blog post plus PDF like [16:39] everything can live inside of one [16:41] notebook. We can just have example. [16:43] Okay. So I want you to add article um [16:47] about cloth code um and notebook about [16:51] cloth code and obsidian uh search online [16:54] and add three articles and um add some [17:00] markdown file from our obsidian base [17:02] which is related to to this. Now let's [17:05] go back to notebookm and u in the [17:08] sources we should see those appearing [17:10] right now we have like those videos and [17:13] cloth is finding a new new sources for [17:17] us so those one web search and those one [17:22] we're trying to find the relevant uh [17:24] files in our obsidian vault okay I'll [17:27] just tell okay okay just use grap okay [17:29] okay that's much faster okay we found [17:32] the three articles from the web And we [17:35] added my setup uh guide. Um all right. [17:39] And we are right now adding in real time [17:41] a new sources which you found in the web [17:44] and those are real sources. Um those are [17:47] actually real sources. You can go there [17:50] and go to this website and it's real and [17:54] exactly using cloud code with obsidian. [17:56] Nice. That's what we want. Let me go [17:59] back here. We all saw that cloud.md [18:02] right. uh and added cloud code absion [18:06] course outline. Now you can ask about uh [18:09] like those those questions about uh like [18:12] new sources and that's truly exciting. [18:15] So you can actually upload [18:18] everything PDFs, papers, blog posts all [18:21] into notebook and actually ask questions [18:25] about them. And not lam is very very [18:28] flexible. It supports uh up to 300 [18:32] sources. And today I just uploaded uh my [18:36] daily notes uh 300 um daily notes and [18:41] here are all of them and you can chat [18:43] with them. They're just incredible. You [18:45] can extract different patterns. That's [18:47] just one of the use cases you can think [18:49] about. Additionally, you can also [18:51] configure persona on how not can [18:54] respond. Um I want to configure a [18:57] persona which should be just delivering [19:00] um concise responses and I want you to [19:03] test it just here in the chat just ask [19:06] some question to test this first set [19:09] persona. [19:11] Okay, cool. And right now we are setting [19:13] the persona and like deliver concise [19:15] direct responses, no filler, no [19:16] preamble. And we are asking four like [19:19] three questions in parallel. We just set [19:22] a context for this chat for this [19:25] notebook. And I believe that okay. So [19:29] let me get back to our previous uh [19:32] notebook. So that's the one. Um uh and [19:36] here in the configured chat we see that [19:38] it's appearing and it said response [19:41] links to shorter. That's nice. Um and [19:44] you can actually see the answer is is [19:46] very very uh concise. I love that. [19:49] That's very cool. Very very cool. Now we [19:52] want to add these questions back to our [19:55] obsidian. I just want to import them and [19:57] I want to see them in Q&A log. Okay. [20:00] Okay, so here we need to actually use uh [20:02] flag JSON to get references with uh [20:05] citation resolution. We just like do it [20:07] once again and we're going to get uh it [20:11] here in our obsidian wall synced. Okay, [20:13] we are now extracting the citations from [20:17] all of the questions. We are resolving [20:20] the citations and right now we are [20:23] seeing that they're appearing in our [20:25] obsidian all of those four questions. [20:28] And here is a concise answer. It's very [20:31] very nice. And uh this this question was [20:34] about TZ context management and it makes [20:37] sense that it pick up everything from [20:41] her video. Nice. So the answer is keep [20:45] global instruction short. Uh-huh. I do [20:49] agree with that. And here is advice from [20:52] her to keep it short because it goes [20:54] into the context every single time you [20:57] chat with CL. Makes sense. So it works [21:00] perfectly. Now moving on. So for the [21:03] citations [21:05] um just a diagram right here is that we [21:10] have those um answers and uh they are [21:13] appearing in notebook lm as here. Those [21:16] are the chunks and we can click here. [21:19] You can actually click here like 39 for [21:22] example and we can go to this chunk and [21:24] it shows as this video and exact [21:28] citation from the chunk and we just [21:31] imported this into our obsidian. So we [21:33] can work there and we use those vic [21:35] links with mapping to the headers to [21:38] site exact passage. So we can easily [21:42] trace the claims which clo which [21:46] notebook claim makes and get back to the [21:49] source all the time. And here is the [21:51] difference. Now previously we need to [21:53] open the browser like drag files or [21:55] create 20 tabs go to YouTube find maybe [21:59] the articles that's just not scalable [22:02] and not reliable and just so tiring. you [22:04] just end up duck with all of this uh 100 [22:08] tabs which you never process and then [22:10] you just like close everything and you [22:12] just forget about it. So you lose [22:14] citations. Now with that approach you [22:19] programmatically add the sources. You [22:21] have a curated list of sources and you [22:23] can ask cloud to dig deeper maybe find [22:26] like a different one. And now we are not [22:29] using this black box uh from notebook lm [22:32] where we like add sources using this uh [22:36] fast research or deep research. We're [22:37] actually have more control over the [22:39] prompts over the sources which we adding [22:41] to our obsidian. And then uh we import [22:46] this into our obsidian and then we use [22:49] wik links in our vault to back up our [22:51] citations. Now another exciting part is [22:55] audio overview. You can actually get the [22:59] overview of uh of the topic. You can [23:04] actually do a deep research about [23:05] something and then ask clot to grab this [23:09] uh audio uh from the notebook alarm. [23:12] Here is the audio overview button, but [23:15] you don't have to click it. Um, CL can [23:18] actually add it to our obsidian vault. I [23:20] want you to use [23:24] I want you to use um I want to generate [23:27] the audio overview over the notebook [23:30] which we working with and focus on the [23:33] gaps. Okay, that's very nice. and then [23:36] add it to my Obsidian World um as a [23:39] embedded audio file inside of the [23:42] notebook. Okay, right now notebook um [23:45] Claude just went about generating the [23:47] audio to and he added the prompt about [23:51] uh focusing on the gaps and specifying [23:54] format being deep dive. Now it takes a [23:57] bit a long time. Uh we could see that um [24:01] if I refresh our notebook lm. Oh yes it [24:04] right here. So right now it's generating [24:07] the audio overview and once it's done [24:10] clo going to get it and put it into our [24:13] obsidian vault. Well I've already done [24:15] that and actually here um here is a [24:19] podcast which was generated. Um, I can [24:22] actually uh go on my phone and because I [24:26] use Obsidian Sync, I can listen it from [24:28] my phone. All right. So, here is the [24:31] exact same uh podcast right here. I can [24:35] just play it on my phone. [24:37] >> Summarizes progress. That's the intent. [24:39] >> Intent. That word keeps coming up. How [24:41] is that different from just like a to-do [24:42] list? [24:43] >> So, I can listen podcast on the go which [24:45] uh you can generate um and have [24:48] customized podcasts. I mean here is also [24:52] like other stuff which I played with. Um [24:55] you can go much much like beyond that. [24:59] So one thing which I played is Absian [25:01] canvas. You can actually generate uh a [25:04] mind maps from the notebook and that's [25:07] exact mind map which we which we got [25:10] from there. That's another use case. You [25:13] can also do flashc cards like [25:16] essentially what you have here like you [25:19] can do it exactly the same and ex import [25:21] it into your obsidian. So let's just uh [25:24] go there um into here. Here is example [25:27] of the flash cards. There's 58 flash [25:30] cards. Um it's a silly topic. Um but you [25:35] can just uh let's say okay um what is [25:39] the primary goal of project memory? You [25:41] can just click there and it's going to [25:43] be resuming, right? It's going to show [25:45] the answer. You can generate custom [25:48] flash cards and imported your Obsidian [25:50] to study. That's the audio overview [25:53] that's working. You can also create a [25:56] slide decks. Um, which I'm [26:00] not a big fan of here. It doesn't look [26:04] cool to me. Um, because I can't edit [26:07] this. But that's also opportunity if you [26:11] if you like that that's a diagram for [26:13] for podcasts. You can just take your [26:16] notes and generate audio MP3 file which [26:19] you put into your Obsidian world and [26:21] then on the phone you can access it. You [26:23] can do it through Obsidian desktop or [26:26] you can do it through mobile. And here [26:28] are some other cool use cases if you're [26:32] academic. I think that's a huge [26:34] opportunity because you can tell CL okay [26:38] let me open a new tab u okay so what's [26:42] new on archive regarding quantum [26:43] computing with ritberase [26:46] so now clo can actually search uh [26:48] specific articles about the topic of [26:51] your research and you can actually feed [26:53] back those into notebook on them and get [26:55] a grounded citations I know that it's [26:57] very important for academics so that [27:00] responses are not made up and you have [27:02] full control for full full observability [27:05] over what's going on. And here is the [27:08] actual link to archive paper. I can [27:11] click there and clone just found this [27:15] article and I can provide specific [27:17] directions you know maybe the articles [27:19] are from you know from uh 2010 if I'm [27:24] interested in more uh you know uh [27:25] foundational research about grim race [27:28] all of that you can just import into [27:30] notebook you can also like do market [27:33] research uh for like feeding uh [27:35] competitors blog posts YouTube videos [27:38] you can actually like this is is a [27:41] killer one. I think you can just set up [27:43] your uh company knowledge base in [27:46] notebook. [27:48] I think it's a huge opportunity [27:51] because you don't need like this complex [27:53] rug infrastructure for that. You can [27:55] just put everything into notebook and [27:58] handles everything for you. You can [28:00] imagine that okay you have like this on [28:02] boarding procedure and people have [28:04] naturally questions and you can put the [28:06] docs for your company into notebook this [28:10] way and then when a new employee comes [28:13] in they have a bunch of questions then [28:16] they can just actually ask those [28:19] questions like from cloc from nolm for [28:23] the basic one right and then you don't [28:25] have to spend time and attention of [28:28] other people who need to um let's say [28:31] like answer those basic questions. [28:33] That's one of the opportunity a shared [28:35] Vicki for for your company. Another [28:38] idea. Okay. So, you have this like [28:40] podcast, right? Like Lex Freedman [28:43] episodes like four four hours. You can [28:46] get a bunch of this I don't know and [28:49] maybe you know research about what's [28:51] going on in the field of AI recently if [28:53] you're interested in that. here are all [28:56] of the just use cases and it's all just [28:58] scratching the surface of what's what's [29:00] possible right now with this system now [29:03] and this kind of closes this loop where [29:06] like your nodes your sources go into a [29:09] notebook and then a knowledge graphs [29:11] come back to you which you can explore [29:12] in your obsidian and I believe that's [29:15] highest leverage where you have all the [29:17] sources all the citations connected and [29:21] previously the nom outputs were just [29:23] trapped in the browser right we need to [29:24] copy paste into somewhere into like [29:27] Google Keep into like maybe Absidian [29:29] whatever system you use and it's all [29:31] scattered and there is no consistency [29:33] how you capture that now you can [29:35] actually have like consistent dashboard [29:37] for research topics you want to like [29:40] know deliver and then what's even more [29:44] exciting because it's all your data now [29:47] it's all sits in your computer as a just [29:50] a modern file and you can tell CL Okay. [29:54] Uh let's say I just like want to like [29:56] engage more, ask more about like all of [29:59] the sources which we have here. Like [30:02] it's all here. It's all like saved right [30:05] here. And I can go back to this video. I [30:08] can like ask more questions about this. [30:11] It's truly amazing what's possible right [30:13] now. And here is a full loop. And here [30:17] uh how it works for me how I I think in [30:19] terms of dashboards I have this like [30:21] note called dashboard like this clo x [30:27] obsidian research it's a type of a [30:29] dashboard and I just put there all the [30:32] information so I can clearly see what's [30:34] going on what are the sources what are [30:36] the Q&A and clo can actually read this [30:40] dashboard as well using a recently [30:43] introduced um obsidian cla Right. Here's [30:46] a demonstration. Uh let's say just want [30:49] to and um let's let's start a new [30:52] conversation to demonstrate this. I want [30:54] to read this dashboard. [30:57] Um I just want to get the front monitor. [30:59] Yeah, cool. And present results as a [31:02] table. Oh, presents results as a table. [31:07] Okay. Right now it's going to go and and [31:09] get to resolve those queries. And so [31:12] right now the results would come back to [31:14] us. There you go. Here we have like all [31:18] of the like all of the our Q&A and we [31:21] can ask questions about this. So the [31:24] basis return empty. I think it doesn't [31:26] work with basis. So I just I can just [31:28] translate this into data view and it's [31:31] going to work. So this was a dashboard [31:34] and we have those related source files [31:37] which I just showed and then we have [31:38] those Q&A notes which can be just seen [31:42] all at once and then those source files [31:45] they have those topics as the wiki links [31:48] like such as cloud code second brain [31:50] cast and that's the graph view which I [31:53] showed here those are actually the [31:55] dashboards which we see here and those [31:58] are all of your videos we can add a new [32:01] group here um called notebook lm source [32:05] um uh and right now it's in red so we [32:09] can clearly see and like and understand [32:11] what's happening here. We can also add [32:14] uh new Yeah, we can also color the [32:16] topics in in a different color. I think [32:19] it doesn't work that well, but I'm sure [32:21] that's possible to do. Now, you have [32:23] this like nice color graph whereas the [32:26] red points is your notebook LM sources. [32:29] Those are videos and you can imagine you [32:31] can have like articles, you know, [32:33] different type of sources and those are [32:36] your topics. They are not colored right [32:38] now, but you can make them colored. [32:40] Cool. That's kind of the whole system [32:42] and that's a beautiful graph. Um yeah, I [32:46] think we're coming to the end and that's [32:48] um like notebook uh LM PI project and [32:52] the cloud code skill. So I built um so [32:55] you can just grab a link and in the [32:58] description and start today. Yeah. And [33:01] uh if you want to go deeper, I'm running [33:02] the workshop this Wednesday where I [33:04] build this in your vault. Yeah. And [33:07] that's been um integration of clo code [33:09] and obsidian and notm. And I also build [33:12] the clo code skill. You can just go grab [33:15] there in description. And if you want to [33:18] go deeper, I'm running a workshop where [33:21] we explore [33:23] the Obsidian CLI and clot code [33:27] integration. And here is a program you [33:30] can join for free. Yeah. And uh if it [33:34] was useful, yeah, subscribe. Um, I'm [33:36] shipping uh one of these every week, so [33:40] you can just grab uh tips, tricks, and [33:43] adapted to your workflow. There you go. [33:46] and I'll see you in the next </details>