More Intuitive UI for FastGPT queries

giga

Expandable Search Box:
- Entailment: The search input field would dynamically expand to accommodate multiple lines of text when the character length of the query exceeds the current single-line limit.
- Purpose: This feature is intended to make longer user queries more legible and easier to proofread and edit before submission.
- Impact on Workflow/UX: This would directly improve the usability of the input field, especially on mobile devices where the current character limit before text is cut off from view (observed around 27 characters in a quick test) makes managing longer queries more difficult to handle. Users currently have to manually move the cursor to view the beginning of their query. An expandable box would allow the entire query to be visible.
Query Clear Functionality:
- Entailment: Implement a mechanism to easily clear the search box after a query has been sent. This could be either an explicit 'CLEAR' button or, more intuitively, an automatic clearing of the input field upon query submission.
- Purpose: This feature aims to streamline the process of submitting multiple consecutive queries within the same FastGPT session.
- Impact on Workflow/UX: The current workflow requires the user to manually highlight and delete the previous query before typing a new one. Given that FastGPT does not appear to maintain conversational context between separate queries, this manual deletion step is an unnecessary friction point. Implementing an automatic clear or a dedicated button would significantly reduce the time and effort required to initiate a new search, thereby enhancing the efficiency and fluidity of using FastGPT for multiple information requests.

Users would interact with these proposed features primarily within the FastGPT query input interface.

Expandable Search Box Usage:
- Usage: When a user types a query that is longer than the visible width of the input box, the box would automatically increase its height to display the full text across multiple lines. The user could then easily read their entire query, identify typos, and make edits without needing to scroll horizontally or move the cursor to hidden parts of the text. This is particularly useful for complex or detailed questions.
- Examples: This behavior is standard in many text input areas across various applications, including:
  - Email clients
  - Chat applications
  - Kagi Assistant
- Integration: This feature would simply modify the existing input box's display properties, making it more user-friendly for the current function of entering a query. It extends the usefulness of the input box by making it practical for queries of any length.
Query Clear Functionality Usage:
- Usage: After a user types a query into the FastGPT box and submits it (e.g., by pressing Enter or clicking a submit button), the text they just entered would either:
  - Automatically disappear from the input box, leaving it empty and ready for the next query.
  - Remain in the box, but a clearly visible 'CLEAR' button (perhaps an 'X' icon) would appear, allowing the user to click it to instantly empty the box.
- Examples:
  - Search bars: Many website search bars include a small 'X' icon that appears once text is entered, allowing the user to quickly clear the input field.
  - Command-line interfaces: While not a GUI example, the concept of the prompt being ready for the next command after execution is similar.
  - Some form fields: Input fields that are expected to be used repeatedly might offer a clear option.
- Integration: This feature would add a simple interaction element (either automatic or a button) adjacent to the existing input box or as part of the submission process. It directly enhances the workflow for users who ask multiple questions in a session by eliminating the repetitive manual deletion step, saving time and reducing frustration. Since FastGPT sessions currently function as discrete query-response interactions without carrying context, clearing the box after each query aligns well with its current operational model and prepares the user for the next independent question.

z64

Hiya! Appreciate all the input - I will note however that before long we will be removing the FastGPT page and just replacing it with (redirecting to) Assistant, which is now available to all plans & overall superior in just about every way. So FastGPT will remain as it is until its eventual sunset.

If you would like to try Assistant and give us your feedback on that instead, that would be welcome! Thanks.

giga

z64 Hello! I appreciate the prompt response and information regarding FastGPT's development roadmap.

I am indeed already a frequent user of Kagi Assistant and find it to be a powerful tool in my daily workflows. I began using FastGPT more frequently as of late because I find it to be a great model for efficient web searches. While I think my custom instruction engineering skills are sufficient, and having experimented with creating custom assistants within Kagi Assistant specifically tailored for web search tasks, I found that the FastGPT model performed closer to my ideal. I found it excelled at quickly providing relevant, up-to-date information with clear citations while my custom assistants may have struggled with maintaining one or more of these standards while also staying within the range of cost efficient models.

I completely understand the strategic decision to consolidate functionality and focus development efforts on Kagi Assistant, especially given its broader availability across subscription tiers.

May I then suggest possibly including FastGPT as a proprietary/first party custom assistant like you guys have done with Ki/Code Assistant? A dedicated "Fast Web Search" or similar assistant powered by the FastGPT model could preserve functionality for users who appreciate its specific approach, while still aligning with the plan to centralize features within the Assistant interface.

Thank you again for your time and consideration. I look forward to seeing the continued evolution of Kagi Assistant.

z64

In principle they should not be so different, as they already are using the majority of the same code & infrastructure. I will pass your feedback to our AI team to see if there is something to do to make Assistant fit your use-case / expectations better.

Technically, the "even closer" thing to FastGPT is our Quick Answer, which has the same exact pipeline of search first -> prompt model second, as FastGPT does. I might suggest trying that as a replacement; you can either

Press the Quick Answer button
End your query with a ? to trigger it
Use the !q <question> bang

giga

Ah yes, I have noticed the Quick Answer seems to provide the same quality of outputs I'm looking for in this use case, and I do typically use Quick Answer in most of my standard Kagi Search prompts so I will continue to do so. It is assuring to know it uses the same pipeline so I suppose in that case it's not entirely necessary to integrate the FastGPT model as a custom assistant in the Assistant interface.

I appreciate your informative responses and extend my appreciation to the Kagi team for consistently maintaining these great services.

If there is ever any testing you need done on the user-side for Kagi Assistant or Kagi's AI services in general, I would be happy to help. I am a power user of Kagi Assistant and am experienced with engineering custom models for all sorts of different tasks. While I don't have professional qualifications in this field I have a strong interest in it and would love to help your AI team out in any way I can.

I am an active lurker in the Kagi Discord, username 'gigacapital'. Feel free to reach out anytime for deeper user insights on Kagi AI services.

Thanks again!