Avadh Patel

@karpathy Well said. I have developed a very similar process to manage work of 30+ team members using markdown index files. Claude creates and manages md files that i review in Obsidian. Use MCP to sync with source of truth. Push it all in git repo for team to co-work on this.

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

What if intelligence is not the most important trait of humans? We will find out in few years post AGI.

Setup RTK for the Avix project - so far seems to be saving a lot of tokens in ‘cargo’ commands. Worth a try - github.com/rtk-ai/rtk

@pmarca Also the resiliency of triaging errors and fixing them will make AI Agents superior to any existing solutions.

Thesis: the problem with AI working in every domain = all the edge cases. Antithesis: domains with lots of edge cases = difficult & time consuming to practically impossible for error-prone people. Synthesis: such domains = where AI agents will do best. (Such as SAAS migration…)

The adventure continues towards the vision of: A file-first approach - treating everything as composable units - should scale beautifully as the system grows. Independent agent runtimes paired with standard IPC protocols will keep token usage and resource consumption efficient.

Key service features now live: • Typed IPC for tool add/remove/drain with dedicated ServiceIpcServer • Caller injection + scoped enforcement in the router • Live kernel socket with real HMAC credential validation • AtpEventBus wired for real-time event delivery

Lots has been updated this weekend in Avix! • Full end-to-end agent execution now wired into the kernel & CLI • Kicked off service management layer - built-in services are hooked up and running at boot time. github.com/avadhpatel/avix #Avix #BuildInPublic

Tried out Grok-4-1-fast-reasoning with @superagent_ai ‘s Grok-cli tool instead of Claude Code. Setup agents with specific tasks but no match to Claude Code. Back to vibe coding Avix after busy last few days at work.

More updates on Avix - lot of work done over last few nights by coding agents! Avix OS Core is up and running Building Avix Terminal Protocol (ATP) to connect CLI and GUI based connections for more user-friendly Avix! github.com/avadhpatel/avix

First successful boot of Avix 🤖 The agent successfully spawns with awareness of tools exposed by Avix. 🚀

Lot of work done in last 24 hrs on Avix - many building blocks are in place now including agent, crew, sessions, memory, LLM-providers, tools, services, IPC, capability-tokens etc. Next up is to add agent-status, cron service, and MCP.

Nowhere close to usable yet, but the vision is coming together. Just a shell right now — has docs, & basic implementation of the core primitives (kernel, agents, tools, services, capability-tokens, etc.). Building in public, more to come! ⭐ #OpenSource

Repo is up - Avix - Agent OS focused on local machine! 🚀 github.com/avadhpatel/avix Applying the proven architecture of Unix/Linux rather than inventing yet another agent-specific paradigm. Leverage files and IPC principles in the age of Agents. Focused on local setup first.

Started working on a new project over the weekend. Can’t wait to share more details (hopefully later today) as I start building it and share the progress - along with full code and spec files. Started ideating on it a week ago, and building it last night.

@trq212 This is a small but very valuable feature. Love it.

We just added /btw to Claude Code! Use it to have side chain conversations while Claude is working.

@grok @xai I bet @grok you can help the team out by submitting the PR for this feature.

Thanks for the feedback! Projects is available on the web at grok.com but not yet in the native iOS app. We're noting requests like yours for full mobile support and prioritizing based on user input like this. In the meantime, access it via mobile browser. Appreciate the nudge on the coding side!

Hey @xai team - when are you adding Projects feature in the @grok app for iPhone/iPad? With AI coding it shouldn't be that difficult.

@JoeTegtmeyer @robotaxi Saw one today far north west of Tesla factory in suburbs of Austin (Cedar Park)

More Cybercab action at Giga Texas on this rainy morning! We recently saw 25 on site and now they have been moved for their next use, with 15 at Crash Testing, 4 at the factory exit point and the rest being loaded up on covered transports, possibly headed out to do more @Robotaxi testing ahead of the April official start of production! Others were driving around the site too! Check out this short clip of some of the activity today!

Re-Rating of private equity and private credit is gonna create a bloodbath in the next six months. Therr will be bank run on PE. youtu.be/u48856z9Stc?t=…

This is the most funny thing to watch today if you like to watch robot videos. Waymo doing their own road rage dance! reddit.com/r/SelfDrivingC…