Real-Time Data
Bring the latest information into your conversations
Enable real-time data integration to instantly enhance your conversations with current information, weather updates, sports scores, news headlines, and more.
Real-Time Data can be collected a few different ways. You can use Msty Sidecar or a third-party service such as Jina or SearXNG.
How RTD Works
RTD is not powered by the model itself, but rather by a search engine that retrieves information from the internet. When you enable RTD, Msty will use the configured search engine to fetch the latest data and provide it to the model as context for your conversation.
The model will then use this context to generate responses that are informed by the most recent information available.
How to enable Real-Time Data
Click the globe icon inside the chat input box to enable the Real-Time Data feature. When this feature is enabled, Msty retrieves information from the internet and provides it to the model for enhanced context in your conversations.
Note that some models are better than others with looking for real-time data, including for the current date. You may need to set the date range to specify a date or be explicit in the prompt to include the exact date you are interested in.
Real-Time Data Options
Right-click on the globe icon to configure advanced options, including which method to use for real-time data.
Sidecar
To use Sidecar, you need to have the Sidecar app installed and running.
Sidecar provides the follow advanced options:
- Search Provider can be configured to use Google or Brave search engines
- Custom Search Query allows you to modify the search query to improve the quality of results
- Limit by Domain restricts your search to a specific domain (e.g., nasa.gov)
- Extra Operators allows you to add special search operators, such as
site:orintitle: - Search Date Range allows you to limit the search results to a specific time frame
Jina
You can setup a Jina account and grab an API key. Enter the API key in the Use Jina tab in Msty Studio.
SearXNG
SearXNG is a free internet metasearch engine which aggregates results from various search engines without tracking users. You can use your SearXNG instance to search the web for real-time data.
In Msty Studio, enter your instances endpoint on the Use SearXNG tab. For example, https://searxng.example.com/.
For additional guidance on setting up SearXNG, check out our Youtube video on How to Set Up SearXNG.
Allowing insecure content
Studio uses a JSON API so you will typically need to ensure your instance is being served over HTTPS.
However, some browsers have settings to allow for 'insecure' content served over http for a website. Such as Brave via brave://settings/content/insecureContent?search=additional+content > allow Msty Studio to load insecure content.
Real-Time Data Troubleshooting and Tips
If you're seeing irrelevant responses, the issue may lie with one of the following:
Search engines not returning useful results or no results at all.
- Check the context window size for the model by clicking the Model Parameters icon and try increasing it if possible.
- If no results are returned, try selecting a different search engine. Some search engines will limit, throttle, or block requests that are too frequent or too broad.
- If you have a VPN enabled, it may be affecting the search results. Try disabling the VPN and see if that resolves the issue.
- Try using Jina or SearXNG for more reliable results.
- If using Sidecar, confirm Studio is connected.
Model not using the real-time data effectively.
- Increase the model's context window size if possible as the context window size determines how much information the model can consider when generating responses.
- Ensure the model is capable of understanding and utilizing real-time data. Some models may not be optimized for this purpose.
- If the model is not using the real-time data effectively, try refining your prompts to be more specific about the information you want.
Best Practices
- Use clear, specific prompts that suggest a need for recent or factual data.
- If RTD is enabled but no sources are cited, be cautious: the model may have hallucinated.
- Consider having a system prompt note that a search will occur — it may improve result consistency.
- If zero results are returned, have the model return an error message or fallback notice rather than hallucinating a confident but false answer.
