New features and improvements made in Braina version 3.0 are as follows:
- Deep web search engine for LLM (Use with both local and cloud LLMs)
The deep search feature searches the web in depth to find more context related to your query, which can be helpful for research.
You can enable deep search option from the status bar present at the bottom of Braina’s window. You can also enable option for third party search providers from settings -> search tab for better web search results. Using external search providers can enhance your results, though it might add an extra 5 to 10 seconds to the search process.
Please note that deep search passes a large amount of context to the LLM. If you are using a cloud-based LLM, this may result in higher credits usage.
If you are using a local LLM, make sure it has a context length set to 8192 tokens or more for deep search to work properly. If you have less powerful hardware and cannot set the context length higher than 8192, it is advisable to keep the input context option as ‘None’ or ‘Previous Interaction’ when using deep search with a local LLM. If your hardware limitations do not allow you to set the context length higher than 4096 tokens, it is advisable to disable deep search and just use normal web search. However, you can enable the third-party search providers option if needed for better results.
Tip: Use following LLM system prompt while using web search:
Use provided web results to give helpful answers along with reference links. Current date and time is {timestamp} {timezone}
Learn more: 10 Use Cases for using Web Search with LLM
- Screenshot attachments for vision LLMs
Screenshots have become an integral part of our digital communication, and when combined with Vision Language Models (VLMs), they open up numerous possibilities for automated analysis and assistance. Braina provides an easy way to add selections and desktop screenshots as attachments to multimodal LLMs. To attach a screenshot, click on the clip icon (on the left side of input field) and select “Add Selection screenshot” (Ctrl+H) or “Add Desktop Screenshot” (Ctrl+J). To take screenshot from other monitor screen, move your mouse pointer to that screen and press the keyboard shortcut (Ctrl+H or Ctrl+J) while the focus is still on Braina’s window.
- New LLM/Advanced AI Engine: Pixtral Large
Pixtral is a multimodal LLM from Mistral AI that is built on top of Mistral Large 2. - New advanced AI Image Engine: Flux 1.1 Pro Raw
Flux 1.1 Pro Raw model captures the genuine feel of candid photography. Using this image engine, you can generate images with a less synthetic, more natural aesthetic. Compared to other text-to-image models, raw mode significantly increases diversity in human subjects and enhances the realism of nature photography. - Local LLM support for Llama 3.2 Vision
The Llama 3.2-Vision instruction-tuned models are optimized for visual recognition, image reasoning, captioning, and answering general questions about an image. The models outperform many of the available open source and closed multimodal models on common industry benchmarks. - Template placeholders for LLM system prompt
In the context of designing an LLM (Large Language Model) system prompt, Braina’s addition of template placeholders such as {date}, {time}, {timezone}, {timestamp}, {username}, {firstname}, {lastname}, {city}, {country}, and {area} significantly enhances the system’s ability to provide personalized and contextually relevant responses. These placeholders allow for dynamic content insertion, enabling the LLM to tailor its interactions based on real-time data specific to each user or request. For instance, incorporating {date} and {time} can make greetings more relatable, while using {username} and personal details like {firstname} and {lastname} fosters a sense of familiarity and engagement. Additionally, location-based placeholders such as {city}, {country}, and {area} can help the model provide localized information or recommendations, thereby improving user satisfaction and interaction quality. - Change input field font size
You can now change font size for the input field of Braina from Settings -> Appearance tab. - Message information
Hover your mouse pointer over a message to see information about that message, such as the LLM/AI engine name, message time, etc.
Braina will also temporarily show details such as no. of words in the output message, total time taken (loading + inference + display) and display speed (words per second). Please note that the display speed is in words per second and not tokens per second. The speed is based on total time taken that includes model loading and display time and not just the inference time. - Copy button for code generated by LLM
Now you can easily copy code generated by LLM with a single click using the copy button (appears on the top-right side of the code block as shown in the image below)
You can now also copy text from any message by right clicking on it and selecting “Copy this message” option.
- Option to delete an individual message from the chat conversation
You can now delete individual messages from the chat history by right clicking on the message and selecting “Delete this message” option.
- Minor bug fixes