Show HN: SEC filing AI built by equity analyst and PM

Click on the little chat icon at the bottom right of the screen. It should be visible on most pages on the site.

You have 4 core levels to choose from: off, low, medium, and high. Set the default model on your account page. If you want to use a different model, start your question with the command:

• /off Who is the CEO of Microsoft?
• /med Why does AMD report amortization in COGS?

Use the /help command when chatting to get the full list of commands you can use.

Dbot uses more than 10 different models together in a seamless manner. These models are bucketed in the 4 categories based on their capabilities, their ability to think, context length (how much data they can work with), and speed. Our house-default setting is 'low'. We recommend starting with that and then moving up or down based on needs.

More advanced models typically have more internal knowledge and context lengths that can be used. If you can't get the answer you're looking for, often just moving up one level and asking the same question again will get you a good answer.

For example, if you are using the 'low' intelligence level, try asking the same question with '/med' or '/high' at the start of your question. This will switch the model to a more advanced one that can provide better answers.

Once you select a new intelligence level, the subsequent interactions will continue using that level until you change it or exit the conversation.

Use the '/web' trigger to search the web. A good reason for including web search is to bring in data into your research work that is not included in filings. This includes potentially detailed industry knowledge (e.g. /web How long does it typically take to develop a new drug for hypertension?), product reviews, policy changes, etc.

Once the data is in the conversation context, you can use this context (fancy way of saying chat history) to provide important information to the LLM to generate better answers.

Our coverage is increasing every single day - both in breadth (number of companies) and depth (historical coverage). We are currently focused on providing coverage for US companies in the S&P 500. Coverage will gradually be expanded to other major indices like Nasdaq 100, Russell 1000, etc.

We will soon expand to include US-listed ADRs as well. Additional global markets will follow in due course.

Customers can request companies for inclusion. Please email us with your requests.

Dbot is excellent at diving in to details about companies' filings. You can pull the as-reported financials, discuss details about the various segments, analyze evolution of margins, compare with competitors, etc. If the data is in the filings, Dbot will get it for you. If you notice its struggling, move up to the next higher thinking level to try and get your answer.

If you would like to perform complex analysis, we recommed building up to the ultimate query gradually. If running analysis across multiple companies and multiple time periods, start with one company, build up its history, then proceed to the next. Once you have built up the context, you can then ask the agent to perform the ultimate analysis.

LLMs are still a relatively new technology, so there are always pitfalls. Treat Dbot like a smart junior analyst - the work will most likely be done well, but if your life depends on it, then best to verify. Here are some key aspects to keep in mind:

• We have strived to minimize hallucinations, however we recommend verifying information with the references.
• To ensure the best results, we recommend providing sufficient detail and guidance in the question. The better the framing of the question, the higher the likelihood of the response matching your expectations.
• Things will likely blow up if you ask it to run very complicated analysis up front. Particularly challenging scenarios are where questions require searching and analyzing multiple companies and multipe time periods in one shot. We recommend building up to complex topics in a step-by-step fashion.
• Complex queries are best attempted at the 'high' intelligence levels.
• If you notice the agent is going down the wrong rabbit hole, you can use /stop to halt execution.
• You can also use /clear to clear the internal working memory. This can help to reset the agent's train of thought without losing your existing chat context.