GitHub user bobbai00 created a discussion: Proposal: Ditch Angular and Switch 
to a new frontend framework

# Proposal: Ditch Angular, Switch to React

## Today's Texera Frontend

Texera's frontend is built with Angular 21. It has three major components:

1. **Resource pages**: the dashboard and the community hub list workflows, 
datasets, models, and computing units, with detail pages for workflows, 
datasets, and models. They have 45 components built on ng-zorro (Ant Design for 
Angular).
2. **Collaborative workflow canvas**: the workflow editor is built on JointJS 
3.5, and users edit the same workflow together through Yjs over `y-websocket`. 
About 2,100 lines of glue code connect JointJS to the workflow model.
3. **AI components**: the agent chat panel is hand-written. It has about 5,300 
lines of code, including our own WebSocket protocol, message rendering, and 
tool-call views.

## Limitations

Based on my experience of developing several components in Texera, I found 
three major limitations with current Texera's frontend framework:

- **Limitation 1: Limited community support for AI components**: Angular is no 
longer a popular frontend framework. For example, React has 30 times more 
weekly downloads (252M vs. 8.4M) and 2.5 times more developers (44.7% vs. 
18.2%). As a result, many new frontend components, especially advanced AI 
components, support only React and not Angular.

- **Limitation 2: Deprecated workflow canvas library**: our workflow canvas is 
built on JointJS 3.5, a 2022 version whose npm package is now deprecated. Newer 
libraries, such as React Flow, are much faster in our tests. Unfortunately, 
React Flow supports only React, so we cannot adopt it in Angular.

- **Limitation 3: Ad-hoc and duplicate code**** for each resource**: each kind 
of resource (e.g. workflow, dataset) has its own hand-built pages, which are 
often copied from one another. For example, the model detail page repeats 65% 
of the dataset detail page's template and 88% of its styles. Moreover, these 
pages hard-code 210 colors instead of using one theme, so they do not look 
alike. As a result, every new kind of resource, such as agents, needs yet 
another copy.

## Proposal: Switch To A New Frontend Framework

To solve these limitations, I suggest switching to a new frontend framework. To 
identify an eligible framework, I think three requirements need to be met

- **Requirement 1: Large developer community**: the framework should be widely 
used, so that new libraries, fixes, and developers are easy to find.

- **Requirement 2: Mature AI components**: the framework should have mature 
libraries for AI components, such as chat, so that Texera does not need to 
build them from scratch.

- **Requirement 3: Real-time collaborative workflow canvas**: the framework 
should have a mature node-graph library that works with shared editing, so that 
several users can edit the same workflow together.

I compare today's Angular with four candidates against these requirements: 
React, Vue, Svelte, and Solid.

### Candidate Comparison

| Requirement | Angular (current) | React | Vue | Svelte | Solid |
| --- | --- | --- | --- | --- | --- |
| R1. Large developer community | 18.2% of developers: 8.4M weekly downloads | 
**44.7% of developers**: 252M weekly downloads | 17.6% of developers: 25M 
weekly downloads | 7.2% of developers: 9.4M weekly downloads | Not in the 
survey: 11.2M weekly downloads, largely via TanStack devtools |
| R2. Mature AI components | Early: 
[CopilotKit](https://www.npmjs.com/package/@copilotkit/angular) 0.6, [AI SDK 
binding](https://www.npmjs.com/package/@ai-sdk/angular) | **Mature**: 
[assistant-ui](https://www.assistant-ui.com/docs), 
[CopilotKit](https://www.npmjs.com/package/@copilotkit/react-core), [AI 
Elements](https://www.npmjs.com/package/ai-elements) | Good: [Nuxt UI 
chat](https://ui.nuxt.com/docs/components/chat-messages), 
[CopilotKit](https://www.npmjs.com/package/@copilotkit/vue), [AI SDK 
binding](https://www.npmjs.com/package/@ai-sdk/vue) | Partial: [AI SDK 
binding](https://www.npmjs.com/package/@ai-sdk/svelte), [AI Elements 
port](https://github.com/SikandarJODD/ai-elements) (community) | Weak: [AI SDK 
binding](https://www.npmjs.com/package/@ai-sdk/solid) deprecated, no chat kit |
| R3. Real-time collaborative workflow canvas | Deprecated: [JointJS 
3.5](https://www.npmjs.com/package/jointjs); alternative [Foblex 
Flow](https://www.npmjs.com/package/@foblex/flow), 98k weekly downloads | **24M 
weekly downloads**: [React Flow](https://reactflow.dev/examples), [Yjs 
example](https://reactflow.dev/examples/interaction/collaborative), [Hocuspocus 
binding](https://www.npmjs.com/package/@hocuspocus/provider-react) | 1.3M 
weekly downloads: [Vue Flow](https://vueflow.dev), used by n8n, no release 
since Jan 2026 | 358k weekly downloads: [Svelte 
Flow](https://svelteflow.dev/examples), no collaboration example | 1.6k weekly 
downloads: [solid-flow](https://www.npmjs.com/package/@dschz/solid-flow), 
community port |
| Meets all requirements | No: misses R2 and R3 | **Yes** | Yes, weaker on R3 
compared to ReAct | No: misses R2 | No: misses R2 and R3 |

> Developer usage is from the [Stack Overflow Developer Survey 
> 2025](https://survey.stackoverflow.co/2025/technology#1-web-frameworks-and-technologies).

## My Proposal: Switch to React

Based on the comparison result, I propose switching to** React**, because it is 
the only candidate that meets all three requirements.

This is apparently a huge topic. I hope developers, especially those who are 
familiar with Texera's frontend development, can chime in.


GitHub link: https://github.com/apache/texera/discussions/9387

----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]

Reply via email to