excited to share a walkthrough of @arsumbrisai build knowledge bases in typed markdown with your agents via mcp, then build your own apps and plugins on top of it everything lives in repos you can share and extend. a bit like packages for knowledge work

agentic note-taking is building a markdown vault together with your agent and now you can build the tools, skills and apps to work with those notes in the same workspace

this is how i browse my llm wikis

just build apps on top of your markdown vault > a todo app for your tasks > a kanban board for your projects > a crm for your customer notes your notes and the apps you build around them all live in the same workspace

how to build you first knowledge graph in arsumbris: 1) create a new tree research workspace 2) start a new agent session 3) invoke /research-premise to define your research goal you can start with something simple like: “/research-premise i want to research the zettelkasten method” you can get more specific, but the skill will help you work out the premise 4) the skill also creates some starting questions as typed files you can inspect them in the file tree 5) if youre happy with the questions, tell your agent: “lets use subagents to research each question in a separate context window. invoke the /research skill in every subagent” 6) once the agents finish, read the articles and adjust them as you like for example, you can ask your agent to add au-writing-style to your workspace reload the app with cmd + r, then ask: “lets use /apply-writing-style on those research articles” 7) use /mediate to explore further it checks the articles against your premise and proposes follow-up questions you can repeat /research → /mediate for as long as you want to keep exploring 8) when you have enough source material to start building a graph, ask your agent to add au-weave to your workspace 9) use /weave-premise to define what you want to extract and build if you dont have specific preferences, start with the default extraction template: “/weave-premise i want to build a wiki about the zettelkasten method using the default extraction template from au-weave” 10) start with one source: “lets take the first research article and apply /mine” (btw i did not play with jev yet but build own experiments and lmk if its nice) 11) this gives you a raw extraction of candidate objects from the article follow up with: “lets apply /cut” it evaluates each candidate against your premise and whats already in the graph 12) weave the selected candidates: “lets use subagents to /weave the selected candidates from this extraction” 13) follow up with /reweave to connect the new nodes into the existing graph and fold in enrichments 14) occasionally use /tend to check the graph for issues and maintain it 15) repeat for the remaining sources you can adapt the types and workflows as you go, or add packages like au-govern for audits but this should give you a first experiment to work through

you can also watch it on yt: youtu.be/UaldcPUthU8

in retrospect i cant believe i wrote knowledge graphs and second brains by hand

i dont think post-training will make md files redundant i want to be able to read what my agent is working with and check if it makes sense (legibility) model weights are a black box. you cant read through them and check what the model has learned think about all the decisions in a company and how they depend on each other when you work with other people (or agents), you want those decisions and relationships written down somewhere so you can check them and agree on what they mean of course some procedural skills may become less useful. but you still need a surface where you map out what you know in a way thats verifiable for you pretty sure markdown will stay because its a format that both humans and agents can read, inspect and edit we just need to stop writing slop into md files context needs to be highly curated and checked against your actual expectations and understanding linked and TYPED markdown files let you model your knowledge / understanding in a graph damn it, just stop writing slop and start modeling your knowledge

codex is down rn and claude runs past the session limit happy for you...

a 22-minute technical deep dive into our agent framework and knowledge IDE

we keep trying to solve agent memory, but what we actually need is a shared understanding

ehm guys... there is a 3d capybara pet in my IDE @feriederich said everybody should design their own pet, what the hell is going on?