GRAYBOOK LABS
Graybook Labs
We build advanced applications on custom LLMs, retrieval-augmented generation and natural language processing, and put them to work where the numbers matter most: trading.
Vig, by Graybook Labs
Vig gathers the right context from your trades, positions and research, does the math in code, and explains the result in plain language. Here’s how each question goes, end to end.
VigAI quantExample trades. The math is real; the trades are made up.
Vig gathers the data behind a trade, calls its tools to fetch prices and compute the numbers, then answers in plain English with the working shown.
Three disciplines, one goal: software that reads, reasons and explains as well as a sharp analyst, at machine speed.
General chatbots don’t know your instruments, your rules or your vocabulary. We pre-train, fine-tune and align models on the data that matters to you, then shrink them until they run fast on hardware you control.
Learn moreRetrieval-augmented generation lets a model answer from your documents instead of its memory. We design the whole pipeline, from parsing and chunking to hybrid search, reranking and citations, so every answer points back to where it came from.
Learn moreNews, filings, central-bank statements and chat logs are full of signal and noise. We build NLP pipelines that read them at machine speed: pulling out entities and numbers, tagging events and sentiment, and handing clean data to the systems that trade on it.
Learn moreVig, our AI quant
Vig works out the probabilities, the sizes and the greeks, then tells you what they mean in words that make sense. Ask why a trade went wrong, whether a system holds up or what a number on your screen is really saying.
VigAI quantWhy did my NQ stop get hit right before the move went my way?
Mostly noise. In the 15 minutes after your entry, NQ’s 1-minute moves had a standard deviation of about 4.2 points. Treat price as a random walk and the chance it touches a stop 12 points away in that window is about 46%, before your idea has had time to work.
A 24-point stop on 1 contract risks the same $480 as 12 points on 2, and cuts the odds of being shaken out to about 14%.
Example conversation. The math is real; the trades are made up.
Language for the explanation, code for the math, retrieval for the facts. Each part does what it’s best at.
Your question pulls the right context: trade history, positions, market data, contract specs and research.
A model tuned on trading text works out what you’re really asking and what it needs to answer it.
The numbers are calculated, not guessed: probabilities, sizes and greeks come from code that runs.
The answer comes back in plain language, with the working shown and the sources cited.
We don’t just call an API. We train, fine-tune and serve models ourselves, with the frameworks the frontier labs use.
Every model we build is trained, fine-tuned and evaluated in PyTorch, from LoRA adapters on a single GPU to full fine-tunes across a cluster.
We fine-tune and serve open-weight DeepSeek models, including mixture-of-experts reasoning models, on infrastructure you control.
For the largest runs we use NVIDIA’s Megatron-LM, with tensor, pipeline and data parallelism, to train models too big for any one GPU.
The hardware, clouds and platforms we build on
//@gbs 1// A trailing stop that follows price by a multiple of the ATR and flips when price crosses it.indicator("ATR Trailing Stop", overlay = true, category = "Volatility") input int length = 14, "ATR length", min = 1, max = 200input float mult = 3.0, "Multiple", min = 0.5, max = 10 series float offset = atr(length) * multvar float stop = navar int trend = 1 if na(stop) { stop = close - offset} else if trend == 1 { stop = max(stop, close - offset) if close < stop { trend = -1 stop = close + offset }} else { stop = min(stop, close + offset) if close > stop { trend = 1 stop = close - offset }} plot(stop, "Stop", color = trend == 1 ? #26a69a : #ef5350, width = 2, style = step)Tell us about the data and the question. We’ll tell you honestly whether a model can answer it.