Sam Partee

@svonava @synopsi @hasp_dev 🫡

@synopsi @hasp_dev @SamPartee works on auth for agents, maybe yall should chat ^_^

It is clear to me that we will have to rebuild the whole security suite for agents: - Password manager - Firewall - Antivirus I got first covered (@hasp_dev), working on the second and thinking about changing mdgrok into the 3rd.

In honor of @LangChain Interrupt today, here's my favorite Agent I made with Fleet + @TryArcade Don Forgettabouit - a very sarcastic reminder agent Started as a simple email triage agent I setup in 5 minutes and now it's the first email I open everyday. Today's gem - "Did you miss the 'Action Required Today' part of that email from 3 days ago?"

We just open-sourced Agent Library, a local-first memory layer for AI agents built by @SamPartee. Your agent's memory should be something you can hold in your hand. A file you own, diff, and roll back. More on the blog: arcade.dev/blog/agent-lib…

S/o to @torresmateo for pushing me to release this! agent-library is just as it sounds - a library for an agent. It started as my personal obsidian connector for Arcade, but has grown into a multi-modal knowledge management system. Tweet thread on the design and what I learned building it very soon.

@BraceSproul @cognition What did you make this demo on?

Fleet can now generate and render SVGs and Mermaid diagrams inline! Ask it to make you a diagram, and watch it work its magic. I'm using a mix of our GitHub tools, and @cognition's DeepWiki to make the diagram in this demo. Works super well!

@jacob_posel I hate when people just do ads but you wrote it for me so i'll just leave this here @tryarcade

We should get rid of API keys, too risky Instead, every time my agent wants to use a tool I should have to give it an MFA code from the Google Authenticator app (via Telegram)

@ivanleomk do it yourself, download the HAR file, claude out MCP tools to get the session token and use the same routes as the UI to manipulate the backend without opening one other than to log in.

What’s the best way that people are allowing coding agents to control their browsers

@paulg Most can't conceive the size a pond can be. Furthermore, many choose to balk at the big fish or pond itself because of this. few enjoy the challenge of defeating bigger fish or migrating ponds so to speak.

You don't know how big a fish you are till you try a big pond.

MCP and Skills are not mutually exclusive!!! nor does either have a exclusive locale it's like people forget MCP used to get flak for being "local only". Skills can use MCP tools. MCP will soon (conjecture) support skills as resources as many clients do today.

@mattpocockuk honestly I don't love some of the dev philosophies in here. when tested (go and python codebases) the results were mediocre at best. I do love this language here: github.com/mattpocock/ski…

FYI I just shipped a huge improvement to /improve-codebase-architecture It now ships with a glossary of terminology to describe good/bad codebases Essential reading for anyone wanting to improve their codebases: github.com/mattpocock/ski…

I’ve been using a similar approach through an MCP I call Librarian. Organizes knowledge into sections, books, chapters, etc. pip install agent-library Libr add Proper chunking, code splitting, ocr, image, etc for embedding models all small enough to run locally. RRF vector + text search indices built and updated automatically. Looking at my obsidian graph now looks like truly autonomous knowledge organization

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.

Hit me up if you’re at #MCPDevSummit!

@KirkMarple @johnennis Almost sounds like you need… permissions… for … contextual… access… Muahahhahhh

@johnennis Yeah, I just want a “read everything you want” flag. Like if it’s not changing data, I don’t care.

Claude Code really needs a --safely-skip-permissions flag

@samuelcolvin Who wouldn’t feel like an imposter next to guido

Obviously not talking about me, I'm an imposter, but the rest of these people - they made the modern world.

Find me a conference with a better line-up, I dare you. There are still some tickets available, buy now. pyai.events/speakers Or, even better, if you're a serious open source developer in your free time, email me and we'll give you a free ticket. samuel at pydantic.dev.

@mihail_eric 🫡

FOMO is the most universal feeling I see across developers in the industry. So many new tools every day: which ones do I use? So many new product releases: which are legit? So many new paradigms: which ones give me the highest ROI? That’s why I’m excited to speak at the Coding Agents conference on March 3rd organized by the MLOps community. I’ll be sharing the stage with product, technical leaders and friends (@TomasReimers , @zachlloydtweets , @SamPartee) from @cursor_ai , @AnthropicAI , @warpdotdev , Arcade.dev. Every talk will be about making coding agents work in production. All signal, no Twitter hype. Hope to see you there! Event link in comments.

@kepano People really don’t know how powerful this combo is. I made one for this but this’ll be way better

1. install Obsidian 1.12 2. enable CLI 3. now OpenClaw, OpenCode, Claude Code, Codex, or any other agent can use Obsidian

@alpic_ai @OpenAI @github @TryArcade @rachelnabors I see you!

We wrapped this edition of MCP Connect Day with 3 fantastic talks from @OpenAI, @github and @TryArcade Thanks to all our speakers for sharing their MCP stories and insights!

@PongInit @openclaw @TryArcade its a move. the new M chips are stellar for local models too.

@SamPartee @openclaw @TryArcade I think this has to be the way. I’ve got a Mac Studio on the way

@openclaw ... but give it everything! Wrote a plugin to give OpenClaw / Moltbot / Clawdbot access to all of @TryArcade tools. Now you can use 100s of services from any of your msg apps i.e. Discord, Telegram, etc. Auth included 👇Me using GCal and Slack from Telegram

@PongInit @openclaw @TryArcade mac pro dedicated.

@SamPartee @openclaw @TryArcade Curious - are you dedicating a machine to OpenClaw to be “always on”? Or running it on VPS?