ai-chatbot
ai-chatbot copied to clipboard
v2 incoming - feedback wanted (updated)
description
we've been working on some exciting updates to the chatbot and would love your feedback on what features you'd like to see next!
we will be doing a complete revamp for the next coming days and be releasing next week
currently in progress
- improving ai elements component library implementation
- database change
- auth provider changes
planned features
- ai gateway multiple models
- folder / chat structure
- ui updates with a nice revamp
what would you like to see?
please comment below with features you'd like to see added:
- new ai model integrations?
- specific ui/ux improvements?
- performance enhancements?
- developer experience improvements?
- other ideas?
- comment anything
your feedback helps us prioritize what to build next!
Mcp support could be nice
Support for different databases and auth providers would be a huge win.
IIRC, using the template today requires setting up a blob store, a Neon database, and using Grok. It would be nice to have some flexibility there.
The biggest blocker for me in using the chat SDK template in the past was the lack of BetterAuth support. Nowadays, that's a deal-breaker for me.
Library for generated artifacts
i was just trying how this works, first time here. already found some errors within less than 5 min playing with it:
Bugs
- npm i doesnt work properly
- when asking the chatbot to "createDocument", it opens a popup and starts writing on it, however, this breaks completely the UI, whenever you try to navigate to another chat or click any button, it reopens this popup constantly
- when "createDocument", you have made it so it can have "revisions" but I'm unable to make it the chatbox edit any of these revisions. ex: Edit section 3. Chatbox response: section 3 doesn't exit..
Nice to have features
- Social login
- "Premium" and "Free" acounts with payment service provider like Stripe, account limitations, etc.
- Folders for chats
- "Team" accounts
- Prompt improver (add button next to the send button to improve the current prompt using AI)
In my case, I would like to see the chatbot updated to Tailwind and Shadcn v4. I've created a PR with these changes; it would be great if you could review it.
https://github.com/vercel/ai-chatbot/pull/1134
Also other features I would like to see:
- File support (PDF and others)
It should be easy to create an Artifacts UI , for every use case like editing text, image, audio, video, and store artifacts like pages in a book, to a library, so you may go back in history and it would be awesome to support realtime collaboration on artifacts
Hey, first off, thank you for the fantastic work on this template. I'm using it as a founder and have some feedback from the perspective of someone who's testing it / modifying it, but doesn't have huge technical depth to truly understand the whole codebase rapidly.
Documentation for a Broader Audience
The existing documentation, especially the two architecture diagrams for the Chat SDK, was incredibly helpful for getting up to speed quickly. Expanding on this would be fantastic.
-
For Developers: More detailed guides on common customizations, like adding or removing tools, leveraging artifacts for one or two use cases with explanations on modifications
-
For AI/LLMs: Including the Chat SDK documentation in the Vercel MCP and adding a project-specific LLM.txt file would make it much easier to use coding agents for faster customization.
User Experience Ideas
Here are a few small UI/UX thoughts that came up during use:
-
Persona Selection: A simple dropdown to set the AI's tone or personality in the chat interface.
-
Chat Organization: A way to organize or collapse chats in the sidebar, perhaps with folders.
-
Easier Sharing: A one-click option to copy a chat's shareable URL directly from the sidebar seems more intuitive to me if the template is meant for chat sharing. The current multi-step process (dots -> share -> select visibility? ) feels a bit cumbersome in the sidebar, and doesn't actually lead to any sharing, just visibility rules change.
Bugs I've Encountered
Sharing a few recurring bugs I've run into:
-
Model Selection Dropdown: The model selection doesn't always seem to apply immediately within an active conversation. For example, if I switch from a "reasoning" model back to a standard "chat" model, I still see "thinking" tokens until I refresh the page.
-
Document Creation Glitches: After creating a document, it sometimes reappears when I open unrelated, older conversations.
-
Revisions UI: The revisions feature is a great concept, but applying suggestions or polishing text doesn't seem to result in any actual changes to the document.
Thanks again for building such a solid starting point, it's a blessing
Shared chat context with persistence would remove a lot of prop drilling
Glad to see progress on a great template! I would LOVE to see migration from AI v4 to v5. A lot of people forked well before v5. I attempted a migration and it did Not go well. Some guide that at least starts at v4.x ai with messages parts that describes a migration path to your latest, including artifacts in the DB, would be Most welcome.
Thank you for your work!
Couple of issues I wanted to address:
1- The tool component from the AI SDK seems more suitable for agentic coding sessions, rather than properly displaying the correct information for an artifact. 2- I think it would be great if we could organize the chats into projects, and allow users to search projects and files. 3- The coding artifact could use a total makeover. At the moment, it doesn't even auto-execute the code, and there are no visual cues to guide the user towards it. 4- I think we all understand that it needs easier integration with different auth providers.
Thanks again!
thank you everyone for the feedback i will take all of it into consideration we will be releasing a video soon with the current changes today/tmr and then continue from there
if anyone does make pr's be sure to ping me! i would love to take a look, i have added ai elements and gateway and want to keep adding more so let me know
I've been working with the AI Chatbot as template for a client project for half a year now. It has been an amazing starting point to built rich AI applications with.
The parts I extended heavily focused on tool calls. I think the repo would benefit from having a centralized approach to move artifacts and tool definitions closer together.
ai/
└── tools/
└── weather/
├── components/
│ ├── server.ts
│ └── client.ts
├── schema.ts
└── tool.ts
Some artifacts like the sheet are coupled with multiple tool calls. If it makes sense they could all be within ai/tools/sheet or a shared primitives directory could be added.
This would make copying tools from the Chatbot Repo into a local repo straight forward.
MCP support and stored memory across chats would be amazing! I started building my own tool calls inside the app but I already have an MCP server I could use instead.
Stored knowledge documents or system prompt updates per user could be cool too, but that's also something I could build pretty quickly and not applicable to everyone.
Tailwin
@dancer do you plan to merge it to tailwind v4 in the upcoming updates?
Hi @dancer,
Thank you for this amazing repo and rolling in all the cool changes. Would it be possible to have the artifacts remember the chat history? It looks to me right now that the artifacts don't have access to the session history.
For example, after discussing for 20 messages, I would like to write a report based on our summarized conversation.
Best regards, AT
some stuff i'd love is!
- Improved design: I worked on this a while back: ai-chatbot. I tried to make the UI look cleaner.
- Tailwind v4 support.
- Social auth with Auth.js (should work fine).
- MCP support.
- Folders
other stuff:
- Custom Instructions
- Custom Themes like (https://intern3.chat/)
how about i18n support? i could take this, plan:
- use next-intl library to implement it
- add a button on right top to switch language, it will display the icon by language
- default lanague is en or condition by user broswer language settings or some how
- support en, zh-cn, jp in the first or more(we could discussion this)
We could unpack the button group from each historical dialogue entry in the history sidebar onto the entry itself and truncate long entry titles, making them more convenient to use.
Fix navigation. When a new chat is started, the URL is updated with window.history.replaceState. See https://github.com/vercel/ai-chatbot/blob/7123fbbd1977ca1de5f7550d37fe05871daeb321/components/multimodal-input.tsx#L122
This messes with the navigation, you can't go back to the previous page. Replace with pushState instead, and remember to do this little trick to make sure navigation works smoothly: https://github.com/vercel/ai/blob/main/examples/next/app/chat/%5BchatId%5D/chat.tsx#L55-L59. This one is crucial
Shared chat context with persistence would remove a lot of prop drilling
Agree, here are some docs for it: https://ai-sdk.dev/cookbook/next/use-shared-chat-context
Also, remembering what model was used for each chat would be great!
Hi, does anyone know where I can watch the video that @dancer mentions?
Hi, does anyone know where I can watch the video that @dancer mentions?
thank you everyone for the feedback i will take all of it into consideration we will be releasing a video soon with the current changes today/tmr and then continue from there
if anyone does make pr's be sure to ping me! i would love to take a look, i have added ai elements and gateway and want to keep adding more so let me know
Hi Dancer, any news regarding the video? What's the general plan now, would be nice to give us an update on what's next, thank you so much!
different auth support would be much appreciated, also require mcp support too.
- MCP Support
- Direct implementation of OpenAi Compatible provider configurations that support artifacts
Generally ... but would like more comprehensive integration for artifacts for our users to fork.
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';
const customOpenAI = createOpenAICompatible({
baseURL: 'https://geoff.magma-rpc.com/v1',
name: 'geoff-custom',
});
return customProvider({
languageModels: {
'chat-model': customOpenAI.chatModel('geoff-30b-2:1'),
'chat-model-reasoning': wrapLanguageModel({
model: customOpenAI.chatModel('geoff-30b-2:1'),
middleware: extractReasoningMiddleware({ tagName: 'think' }),
}),
'title-model': customOpenAI.chatModel('geoff-30b-2:1'),
'artifact-model': customOpenAI.chatModel('geoff-30b-2:1'),
},
});
I think components file is not really organized, i noticed some left over codes as well. Potentially unifiying artifacts into a folder where user can define new artifacts/add custom ones easily by adding files to 2 places (components/artifacts/... and lib/artifacts/...).
Right now i think a lot of code pieces are scattered across, creates a steep learning curve for debugging/updating configs.
My ask remains the same. The move to Vercel AI 5 was, understandably, a large departure from everything previous. I've read everything that I am aware of regarding moving to v5, tried a dozen times, and it still doesn't work. I have yet to see a comprehensive migration guide from this project as to how to migrate it and am stuck on a previous fork that's complex. Am I stuck being stuck, or is there a migration guide coming soon for this project? Otherwise, it seems I'll have to rebuild my entire app from scratch from a fresh fork. If I'm not the only one, then I beg for some updates to the project before moving to AI SDK 5 that, for example, at least gives us access to the latest LLM models. Abandoning everything previous without a proper migration guide has left a lot of folks behind.
Moving away from next-auth into better auth 🤌
Not a must But thought I ll share my feedback
- I work on B2B applications mostly with non native english speakers, while the model itself is lang independent there is no translation for the default text
- I also have added a bunch of features similar to perplexity AI but customising with filters and translations is complicated so I use react-admin
- I also use supabase for auth, database, OAuth, edge functions and maybe in future AI inferencing compared to next-auth since its just plug and play,
Hence I suggest having Auth Options which can be enabled depending the env variables and giving users to select other frameworks as well rather than having to use the default admin panel
I think it would be nice to move this project directly into the AI SDK repository so that it stays more in sync with new features and one always has a go-to example of how to use AI SDK.
https://github.com/vercel/ai/discussions/9618