Fede Cardoso

Last night, I had dinner with a legend in market making, Sasha Stoikov, co-author of the Avellaneda-Stoikov paper. I’m very grateful for the opportunity 🚀

@quantymacro I know a lot more that they are just using hummingbot haha

some have asked me about my time in Renaissance Technologies. although I’m retired I can’t really say much due to NDA, but I have an unseen interview snippet from my ex-colleague Nick (hope the kids are doing well mate) that I’m comfortable to share. a lot of alpha in there

@fengtality Bot it like is hot

Check out today's pod to see how I built this agent currently maintaining a grid strategy in the Hyperliquid WTIOIL perp market

The Bot Pod Episode 2: Deep Dive into the Trading Agents Specification x.com/i/broadcasts/1…

We are live on x and YouTube with @fengtality

We started running operations for a customer on fiat pairs… today we are the 10% of BTC-BRL market with only 40k 🔥

@Prince_Canuma @jtdavies What’s the best way to quantize ?

If you quantize the model you have even more memory savings and speeds ups 🚀 Thanks to @jtdavies for testing it out!

@sovren_software @_hummingbot @fengtality 🔥🔥🔥

@_hummingbot @fengtality @cardosofede Nice pace and clear walkthroughs so far

Introducing The Bot Pod — a weekly deep-dive into AI-powered crypto trading from the maintainers of Hummingbot. In this debut episode, @fengtality and @cardosofede unveil Condor, the new open source harness for building autonomous trading agents. Watch them build and deploy a live trading agent from scratch in under an hour.

Bot pod 🔥🔥🔥

We are live on YouTube, X and Twitch

@Prince_Canuma @GoogleDeepMind @TIIuae @IBMResearch @facebook Congrats! One question, now that ollama supports mlx do u recommend to use it or what’s your preferred way to serve?

mlx-vlm v0.4.3 is here 🚀 Day-0 support: 🔥 Gemma 4 (vision, audio, MoE) by @GoogleDeepMind 🦅 Falcon-OCR + Falcon Perception by @TIIuae 🪨 Granite Vision 4.0 by @IBMResearch New models: 🎯 SAM 3.1 with Object Multiplex by @facebook 🔍 RF-DETR detection & segmentation by @roboflow Infra: ⚡ TurboQuant (KV cache compression) 🖥️ CUDA support for vision models (Sam and RF-DETR) Get started today: > uv pip install -U mlx-vlm Leave us a star ⭐️ github.com/Blaizzy/mlx-vlm

Hey! I’m currently doing that with a personal system called Metabrain, that is the orchestrator and has brains specialized on different domains. I have brain memory, brain-project memory, runbooks (instructions for tasks that I usually ask repeatedly like, rebuild the back and re-run it, and knows that uses the mcp of tmux to attach a specific session). There is a knowledge brain that generates this knowledge on demand and also after finishing a conversation all the text is moved to another session that will analyze it, compare to the current knowledge base, update it or create new entries. I’m working on 5 projects inside the system and the project memory is quite useful since it keeps on it references to other knowledge documents making them easily searchable by task context

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.

@_hummingbot Haha what about hummingpod

🔴 The Hummingbot Pod Episode 1: Trading Agents in Condor starting in about 5 hours, today. youtube.com/watch?v=BorpdJ… Join Hummingbot maintainers Mike and Fede for an in-depth look at Condor, the new open source harness for building autonomous trading agents built by Hummingbot Foundation, launching soon. What They will Cover: → What is a Trading Agent? → OpenClaw and limitations for scalable trading operations → How is Condor different from OpenClaw? → The Trading Agent Standard → Building an Agent in Condor

@fengtality Metabrain is going crazy

I really like how Twitter livestreams shows you which areas were watched the most. People found the 2nd half of last Friday's stream more engaging than the first, esp this trading agent part and @cardosofede's MetaBrain personal AI project 🧠

@gusik4ever Check our latest livestream, we are building condor. Market data from everywhere and we have connected +40 exchanges between dex and cex to trade

the fastest growing GitHub repos in finance this week: 1. TauricResearch/TradingAgents (+9.3K ★) multi-agent LLM framework that runs like a trading firm — analysts, researchers, risk managers all debate before a position opens. works with GPT-5, Claude, Grok, Gemini. 2. virattt/ai-hedge-fund (49.6K ★) team of LLM agents that each play a different role: bull, bear, fundamentals, technicals, risk. the closest thing to an actual AI fund on GitHub. 3. NoFxAiOS/nofx (11.2K ★) autonomous AI trading assistant. picks its own models, pulls its own market data, decides when to trade. added safe mode this week. auto-protects positions when AI fails 3+ times consecutively. 4. Jon-Becker/prediction-market-analysis (2.3K ★) largest public dataset of Polymarket + Kalshi trade history. 36GB. researchers are already publishing papers on top of it. 5. pmxt-dev/pmxt (1.2K ★) CCXT but for prediction markets. one API across Polymarket, Kalshi, Limitless, Myriad. active fixes shipping all week. bookmark this and start today.

Live on twitter now!!!

We will be showing this in an hour… get ready!!!!

The Botcamp Cohort 13 info session is now live 🎓 Learn everything about our program for building AI-based algo trading strategies with Hummingbot: • What you'll learn • Cohort structure & curriculum • Strategy certification process • Q&A with the team 📺 youtube.com/watch?v=xDr5Ir… Part 2 premieres tomorrow — live demo of using Condor to build and run AI trading agents from scratch. Enroll: botcamp.xyz/cohorts/cohort…

This cohort + the trading competition will be amazing

🔥🔥🔥

At Hummingbot, we've started teaching our team to use @openclaw as part of how we work. A few things people get wrong about it: — It doesn't burn through your Claude tokens if you set it up right — Security concerns are real but manageable The reality: you stay in control. Fire off ideas, it handles execution — tweets, code, emails, research, all in one place. A $20 Claude plan pays for itself. We're running an internal setup session next week. More to come 👀