{"slug": "one-brain-means-owning-your-organizational-memory", "title": "One Brain Means Owning Your Organizational Memory", "summary": "Robert Overweg, speaking from his AI visual and production studio Leapfrog, described building a \"one brain\" organizational memory system that makes company knowledge, client context, meetings and decisions queryable through agents in natural language. Overweg's stack runs OpenClaw in a sandbox, a GitHub repository for research and company data, Obsidian for local browsing, and Telegram as the conversational surface, and Tessl has published his AI Native DevCon talk as an agent skill. The system targets two knowledge types — company knowledge about AI, agents, skills and evaluation, and creation-pipeline knowledge such as client assets, delivery requirements and meeting notes — so a small team doing high-volume AI visual work can stop hunting for files.", "body_md": "ARTICLE\n\n# One Brain Means Owning Your Organizational Memory\n\nThe frustration behind my talk was very ordinary: \"where is that file, that presentation, that decision, that client detail, or that piece of research we alr...\n\nRobert Overweg\n\nThe frustration behind my talk was very ordinary: \"where is that file, that presentation, that decision, that client detail, or that piece of research we already discussed?\"\n\nAt Leapfrog, we are a small team doing high-volume AI visual and production work for fashion and brand clients. We create digital people, imagery, video, and related content at scale. Because the team is small and the output is large, we cannot spend our time hunting for folders or reconstructing meeting context.\n\nThat is why I have been building what I call one brain for the organization. It is not an enterprise knowledge-management pitch. It is a practical attempt to make company knowledge, research, client context, meetings, and decisions available through agents in the places where we actually work.\n\nThe key idea is simple: if knowledge sits in scattered files, meeting transcripts, and other people's chat windows, the organization cannot use it well. I want as much of that knowledge as possible in our own stack, queryable through natural language, with boundaries around what should be shared.\n\n## Use This Talk As Agent Context\n\nTessl has turned my AI Native DevCon talk into a [skill your agent can use as context](https://tessl.io/registry/ainativedev/aidevcon-2026-ldn/skills/talk-overweg-one-brain-no-filtering). You can also [watch the full recording](https://www.youtube.com/watch?v=rmxRlpi7xN4).\n\n## Two Kinds Of Knowledge Matter\n\nWe optimize for two kinds of knowledge.\n\nThe first is company knowledge: what we are learning about AI, agents, skills, evaluation, production pipelines, and how our own systems work. That includes research, experiments, internal decisions, and the practices we want to keep improving.\n\nThe second is creation-pipeline knowledge: client information, production preferences, asset details, delivery requirements, meeting notes, and the many different formats clients use when they work with us. Some clients work in Miro, some in Figma, some in Keynote, some in other tools. It is hard to force everyone into one workflow, so we interpret what they give us and bring it into our system.\n\nThose two knowledge types support different work. The first helps us learn and improve as a company. The second helps us produce the right work for clients quickly and consistently.\n\n## Start With A Small Stack\n\nThe first version was simple enough that others can try something similar.\n\nI started with OpenClaw in a sandbox, a GitHub repository that holds research and company data, Obsidian on my machine so I can browse locally, and Telegram as the surface where I can talk to the system at any time.\n\nTelegram is useful because it lets me interact in natural language while walking, early in the morning, or late in the evening. I can ask for context, send thoughts, or look up something that would otherwise require digging through files.\n\nObsidian matters because it is fast and local. I prefer it for this kind of vault because I want the knowledge to feel accessible, not like another slow knowledge base that people avoid.\n\nThe point is not that everyone should copy those exact tools. The point is to start with a small stack you own and can improve gradually.\n\n## Natural-Language Search Changes The Work\n\nThe first return you get is a natural-language sparring partner.\n\nInstead of searching for files, I can search for ideas. I can ask, \"What was that Microsoft CI/CD model again?\" and the system can connect the vague memory to the right research. Because it also understands my context, it can ask whether I mean the material for a talk or a specific internal project.\n\nThat is the difference between file retrieval and organizational memory. The system is not only locating a document. It is helping me recover the context around a thought.\n\nThis matters in day-to-day work. If I need to know whether a prototype was for 500 units or 5,000, or which assets came from a client, I do not want to wait until someone else is awake or search through multiple tools. I want to ask the organization's memory.\n\n## A Research Agent Became My Newspaper\n\nThe second part of the system is proactive research.\n\nI have a cron job running through OpenClaw that tracks accounts and keywords related to agentic engineering, skills, production pipelines, and other areas we care about. Each morning, it gives me a curated update. I do not read a general newspaper for this work. This is my lens into the specific world I need to follow.\n\nBut not everything goes into the shared vault. That distinction is important.\n\nI promote items into the vault only after they seem grounded and useful. I do not want to bother the team with every unvalidated research thread I personally find interesting. Promotion is a discipline. The agent can help analyze whether a skill, article, or idea actually adds something to our current setup, but the decision to share it more widely should still be deliberate.\n\nThat keeps the company brain from becoming a junk drawer.\n\n## The Vault Holds Research, Notes, And Connections\n\nInside the vault, we have research files, notes, todos, and links between related material. At the time of the talk, the research structure had roughly 1,200 files, and the setup could handle that without much extra machinery.\n\nI can ask the chat agent how a piece of our system works and get an answer based on the vault. I can receive notifications when meetings have happened and transcripts are available. I can keep track of my own research, but also connect that research to team knowledge when it becomes relevant.\n\nWe also have a second vault for client information. That is where the production side becomes practical. If someone asks which files came from Calvin Klein, or what an art director tends to prefer, the system can surface that context without making someone manually inspect every file and conversation.\n\nThat is the kind of memory that changes production work. It lets us focus on creating rather than rediscovering.\n\n## Record More, But Own The Data\n\nOne of the more philosophical parts of the talk was about recording.\n\nIf a conversation is not digitized, it effectively does not exist for an AI-native workflow. That does not mean everything should be shared with everyone. It means that unrecorded knowledge cannot be used by the system.\n\nI was influenced by the idea that Bridgewater recorded meetings years ago. At the time, that could sound strange. Now, in a world where agents can turn transcripts into useful context, I see much more value in it.\n\nWe use tools like Granola for meeting transcription because it can combine typed notes with a transcript and produce a useful summary. We also ordered Obi, an open-source recorder, because I want ways to capture conversations while still owning the data and sending it to our own servers.\n\nThe goal is not surveillance. The goal is to make useful organizational knowledge available while controlling where it lives and who can access it.\n\n## Boundaries Are Part Of The System\n\nKnowledge sharing gets difficult quickly.\n\nAre you really going to share every meeting with everyone? What if a call contains something that should not leak? What if client data needs to stay separate? What if leadership research should not automatically become company-wide guidance?\n\nThose questions are why the architecture needs boundaries. We separate personal vaults, leadership knowledge, promoted company knowledge, client information, and agent instances. I do not want everything leaking into everything else.\n\nThe system started messy. We had conflicts, loading issues, merge conflicts between content delivery and GitHub, broken scripts, cron jobs that failed, and ongoing maintenance. Once the base setup works, adding new capabilities can still become a serious amount of work.\n\nThis is not a polished enterprise platform yet. It is an internal stack that gives us leverage because we are willing to maintain it.\n\n## The Architecture Can Grow Gradually\n\nThe architecture now includes private vaults connected to GitHub, GBrain from Gary Tan, vector search, keyword search, translation, direct file reads, and separate agents for different jobs. Telegram is one interface. Obsidian is another. We are also exploring graph layers such as Neo4j.\n\nFor OpenClaw search, we combine memory search over preferences, decisions, and past context; vault context by task or time; broader research through GBrain; and direct file reads where needed. Some work can run on cheaper models to control spend.\n\nThe detail matters less than the pattern: keep the stack modular, owned, and bounded. Add retrieval layers only when the simpler setup stops being enough.\n\n## Start Small And Let One Person Suffer First\n\nMy advice is to start small.\n\nDo not roll this out to everyone immediately. Let one person suffer through the setup first. Run it locally. Check what breaks. Understand permissions, security hardening, and what you are allowed to do with the data.\n\nThen add the pieces that give obvious value. A chief-of-staff agent that shaves time off meeting prep or follow-up can make space for people to improve the system. A research agent that curates the world through your lens can help the team learn continuously. A promotion workflow can decide what becomes shared knowledge.\n\nOne brain is not about putting all knowledge into one undifferentiated pile. It is about owning the organizational memory, making it queryable, and deciding deliberately what moves from personal context to team context to company context.\n\nThe full version of this argument was presented at [AI Native DevCon London](https://tessl.io/devcon/). To go deeper, [watch the full recording](https://www.youtube.com/watch?v=rmxRlpi7xN4).\n\nCOPY & SHARE\n\nRobert Overweg\n\nCo-founder of Leapfrog A.l. Leapfrog A.I. helps fashion brands optimise their studio, e-comm and content creation through A.I. Serving some of the worlds biggest fashion brands. Robert has 13+ years of experience in technology, digital transformation and bringing new technology to large enterprises like Heineken global. Prior to this Robert had a part-time art career, resulting in exhibitions at centre pompidou and Seoul media biennial.\n\nREADING\n\n·\n\n0%\n\nCOPY & SHARE\n\nRobert Overweg\n\nCo-founder of Leapfrog A.l. Leapfrog A.I. helps fashion brands optimise their studio, e-comm and content creation through A.I. Serving some of the worlds biggest fashion brands. Robert has 13+ years of experience in technology, digital transformation and bringing new technology to large enterprises like Heineken global. Prior to this Robert had a part-time art career, resulting in exhibitions at centre pompidou and Seoul media biennial.", "url": "https://wpnews.pro/news/one-brain-means-owning-your-organizational-memory", "canonical_source": "https://tessl.io/blog/one-brain-means-owning-your-organizational-memory", "published_at": "2026-09-17 18:51:25+00:00", "updated_at": "2026-10-02 09:38:52.861435+00:00", "lang": "en", "topics": ["ai-agents", "artificial-intelligence", "ai-tools"], "entities": ["Robert Overweg", "Leapfrog", "OpenClaw", "Obsidian", "Telegram", "GitHub", "Tessl", "AI Native DevCon"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/one-brain-means-owning-your-organizational-memory", "markdown": "https://wpnews.pro/news/one-brain-means-owning-your-organizational-memory.md", "text": "https://wpnews.pro/news/one-brain-means-owning-your-organizational-memory.txt", "jsonld": "https://wpnews.pro/news/one-brain-means-owning-your-organizational-memory.jsonld"}}