TrendSpider Sidekick

What is Sidekick

Sidekick is a language model chat tool. Like ChatGPT or Perplexity, in example. However, it's different from them in a few key aspects, which results in a huge difference.

1. It can not access Internet. This means no unchecked data.
2. It has access to all the data we have on TrendSpider. All the data is specifically adapted to formats which work well with language models.
3. It has a vast set of internal prompts we have designed. We have applied a lot of niche expertise to these prompts.
4. If you ask it to do so, it can access to your custom TrendSpider content, like current chart, watch lists, backtest results and so on.

All of that together makes Sidekick an incredibly useful tool.

What it can do

Sidekick can do analysis for you. It can make sense off a lot of data. Here are its raw "capabilities", but remember that the true power comes from combining these together.

1. It can access any data we have, market or alternative. It can fetch real time candles for any time frame (but never can receive streamed tick updates!). It can fetch fundamentals, it can fetch whatever is in Whats Happening Now. It is smart enough to filter data when fetching.
2. It can see your chart (literally see what you're seeing, including your own annotations, indicators and so on).
3. It can get tickers from your watch lists (you will have to specify from which).
4. It can paint basic annotations on your charts (vertical arrows, horizontal lines, horizontal areas).
5. It has access to the entire TrendSpider knowledge base. This means it's capable of answering questions about how to do things on TrendSpider.
6. It can analyze your backtest results, explain them and highlight weak points. It uses a set of rules and heuristics we have designed, applying our expertise and knowledge.
7. It can explain indicators for you (there's a new right-click menu option on indicators for that purpose). Both in terms of "how to use an indicator" and "what this particular indicator means on this particular chart".
8. It can give you tables with data.
9. It can use charts (line chart, pie chart, column chart) to support its answers.
10. Sidekick has no way for triggering trades or accessing your broker account data.

The most powerful cases emerge from combining these things. You can ask a questions like "what's up with my portfolio today", and Sidekick will fetch tickers from the corresponding watch list, then fetch a ton of data from all the data sources we have, and then give you its summary. Oftentimes, it would highlight things you would not even think about.

Sidekick Personas

Think of personas as "who you talk to". Personas define the attitude and general tone of the language model. All the characters would strive for factual correctness, but they would have different ways of communication. Here's a list of personas we've got so far.

• Neutral analyst: this character is neutral. Reasonably friendly, but neutral.
• Warren, the Long Term Investor: this character values long term vision and discourages paying attention to short term noise. It would focus on long term stuff at every possible occasion. It will call you out for being too short-term, too.
• Graham, the Stoic: this character will try to get to the bottom of every question you ask. Likely making you see your own mistakes or gaps in your knowledge.
• Sheldon, the Data Dude: this character prefers to speak less, and instead to give you more tables, charts and raw data.
• Karen, the Skeptic: this character is often mean, yet always constructive. It will never miss a chance to slap you, and it can even be passive-aggressive. But in doing so, it will often highlight things you would not even think of. Some people on our team find it to be the most useful character, precisely for its capability to be critical all the time.
• Wallstreet Dominator, the Redditor: this character follows tone and values of /r/wallstreetbets. In a somewhat exaggerated manner. It would not be against a good YOLO, but again, just like all the other characters, it values factual correctness above everything else.

Custom root prompts in Sidekick

In the Sidekick settings dialog, you can define a custom root prompt. This is a thing which will be invisibly attached to every message you send. The language model will follow what's requested in there, but only to a certain degree — as long as it does not contradict general Sidekick rules and persona settings.

Different personas are going to follow the custom prompts to different degrees. In example, Karen is very likely to ignore virtually anything, except of straight reasonable demands concerning formatting data (like "always provide a chart if there's data" would work for her).

In example, assume your custom prompt is like Use a lot of emojis and icons in every message. I like them colors, and I like pew pew. My favorite color is green.. "Neutral analyst" persona would add icons reasonably. "Wallstreet Dominator" would go wild and add them almost everywhere. Karen would not only ignore than, she also won't miss a chance to slap you with something like I'm NOT going to use a "lot of emojis, that's ridiculous!.

Custom quick prompts

Once you open Sidekick, you normally have 4 large buttons with pre-defined questions. There are 8 built-in questions like that, and their purpose is to make it easier for traders to familiarize themselves with Sidekick. People can see what it can do, and they can see how different the specificity of requests can be.

However, once you become familiar with it, you don't need that any more. Instead of seeing our examples, you can customize the list of these pre-defined questions. The recommended way is to add pre-built questions which are a part of your normal routine. This way, you can just open Sidekick, click on a button with the pre-defined question, and get your analysis. Like, if you happen to like Palantir, then you can have a pre-defined item like Please give me reasons to keep on holding PLTR, tell me lies, just persuade me to hold it for one more day.

Selecting a language model to work with

Different language models (Anthropic Claude, OpenAI GPT, Gemini Pro) behave in a different way. The only way for defining the best fit for your style is experimenting. Claude is generally slower, but it often goes an extra mile to give you the answers. It's not overly verbose in its responses. Claude works well even with vague questions. Gemini is rather more literal, and it's not as good with vague questions as Claude is. OpenAI is more wordy, but it's way faster than both Gemini and Claude. However, all these characteristics can be adjusted to a good degree, using your custom prompt settings.

Note on Sidekick prices and billing

Every TrendSpider customer has 10 free messages per calendar month (that is, from 1st to 31st), disregarding of their plan type or billing cycle. The only way to have more messages per month is to obtain a Sidekick package.

As a customer, you barely have any control over how many tokens would an LLM take to answer your question. A short question like give me a complete overview of WMT would cause Sidekick to load a ton of data, thereby consuming a lot of tokens. Another simple question like explain my chart would not cost nearly as many tokens. In general, use cases are very different, and token consumption is very diverse thereby. We want to provide customers with a no-surprises pricing model, and that's why we base price on messages sent.

Any time you send a message, your balance of "message available" would be decreased by 1 ONLY IF and ONLY WHEN you receive a textual response from an LLM. In example, in the following cases you will not spend a single "message":

1. You ask it "what's up with my portfolio", it then asks you "what's the watchlist which contains your portfolio" and you never respond.
2. You send a message and just never get a response because of a system error.
3. You ask the built-in question of What kind of data can you use? What can you do for me?. This is the only special case like that, this question is free.

In theory, there might be cases when you ask a vague question and get a follow-up question from the LLM in response. In this case you'll be charge 1 message, even though what you have received is not exactly answering your question.

Limitations and common pitfalls

There are a few things Sidekick can not do at the current stage:

1. It can't code indicators for you. You've got to use the dedicated Custom Indicator LLM for that, there's no integration between them yet.
2. In general, it can't "do things for you in the platform". It can't create or run scans, can't create watch lists and so on. With the exception of "creating basic annotations", as explained above.
3. It can't give direct trading advise. At least it's not supposed to (we prohibit that in the configuration, but it's still an LLM).
4. It can't access any of your content except of what's explicitly listed above.
5. It can't assess quality of your trend lines.

Sidekick can't run scans yet. In case if you ask it to do something what actually must be done via a scanner, it's supposed to refuse and explicitly tell you that you need a scan. However, fairly often, it's going to try and fetch a lit of data and give you results. Unless your universe is tiny (<15 tickers), it's almost guaranteed to give you an incomplete result. That is, it's very likely to miss tickers which are actually a match.