Ask Buffett is invest-like's retrieval-augmented Q&A surface. It answers value-investing questions by pulling the most relevant passages from the Berkshire Hathaway shareholder-letter corpus (1998 to 2025), plus Charlie Munger commentary, then citing the source inline so every claim is auditable against the original letter.
Last reviewed:
RAG stands for retrieval-augmented generation. Instead of asking a large language model to recall what Warren Buffett might have said about a topic (which produces plausible but unverifiable text), Ask Buffett retrieves the actual passages from the Berkshire Hathaway shareholder letters that are semantically closest to the question, places them in the model's context, and constrains the model to produce an answer grounded in those retrieved passages with inline citations. The output is what Buffett (and Munger) actually wrote, not what an LLM guesses they might have written.
The corpus is the complete set of Berkshire Hathaway annual shareholder letters from 1998 through 2025, each chunked into multi-paragraph passages with the source letter year preserved as metadata. The letters from 1998 onward are particularly important because they cover the full modern era of Berkshire: the dot-com bubble and crash, the financial crisis, the post-2009 compounding decade, the Apple thesis, the energy transition, and the recent letters discussing succession.
Charlie Munger commentary is folded in where it exists: the Q&A passages from the Berkshire annual meeting transcripts and the Daily Journal meeting transcripts. When the question touches on a topic Munger commented on directly (capital allocation, psychology of misjudgement, banking sector), the retrieval surfaces the Munger material alongside the Buffett letter passage.
The corpus is intentionally bounded. invest-like does not include third-party summaries, books written about Buffett, or interview transcripts where the verbatim text is uncertain. The bound is the load-bearing difference between Ask Buffett and a generic LLM prompted to imitate Buffett's style.
Four steps. The retrieval step is what stops the model from improvising.
The user's natural-language question is encoded into a dense vector using a sentence-embedding model. No keyword match, no fuzzy regex. The vector captures the meaning, not the surface form, so 'how does Buffett think about debt?' and 'what does Warren say about leverage?' resolve into nearby points in the same semantic space.
The system runs an approximate nearest-neighbour search over the indexed shareholder-letter corpus and returns the top-K passages (typically 6 to 10) most semantically similar to the question. Each passage is a few-paragraph excerpt with its source letter year and Munger commentary attached where it exists.
The retrieved passages are placed in the LLM's context window alongside the original question and a system prompt that requires the model to ground every claim in a retrieved passage and to cite the source letter year. The model composes a coherent answer that quotes or paraphrases the corpus rather than improvising.
The final response renders the cited passages inline so the reader can verify each claim against the underlying letter. If a question cannot be answered from the retrieved corpus, the response says so rather than fabricating; that hard guardrail is what distinguishes RAG from generic generation.
A generic LLM prompted with "answer as if you were Warren Buffett" will produce confident, well-written text that sounds like Buffett. It will also fabricate quotes, attribute positions to letters that never said them, and present plausible-sounding reasoning that has no anchor in what the actual investor wrote. The reader cannot tell the difference from the surface text.
Ask Buffett is constrained at the prompt level to reason from the retrieved passages and to cite them. When the corpus does not contain a relevant passage, the response says so rather than guessing. The trade- off is honesty: an answer that cites no letter is clearly labelled as such, and the user can decide whether to trust the generic background reasoning or not.
The citation format renders the source letter year and a short passage excerpt directly under the relevant claim. If Ask Buffett tells you Buffett favours owner-friendly management with buybacks at low prices, the excerpt from the 1999 or 2011 letter making that point shows up next to the claim. The reader can double-click to read the surrounding context in the original letter.
Ask Buffett lives as a section on every ticker page. Open a verdict and scroll to the Ask Buffett surface, then ask any question about how Buffett would think about this company.
Educational only. invest-like is not a registered investment adviser; Ask Buffett retrieves and cites the public Berkshire Hathaway shareholder letters but does not represent the actual current views of any individual. Always do your own research and consider your individual circumstances.