Skip to content

feat/ai-adapter - #1334

Open
Aryan-Verma-999 wants to merge 8 commits into
RocketChat:developfrom
Aryan-Verma-999:feat/ai-adapter
Open

feat/ai-adapter#1334
Aryan-Verma-999 wants to merge 8 commits into
RocketChat:developfrom
Aryan-Verma-999:feat/ai-adapter

Conversation

@Aryan-Verma-999

@Aryan-Verma-999 Aryan-Verma-999 commented Jun 25, 2026

Copy link
Copy Markdown
Contributor

Pluggable AI Adapter Layer

Adds a zero-dependency @embeddedchat/ai-adapter package that lets host applications inject any LLM provider (Ollama, OpenAI, Gemini, or custom) into EmbeddedChat without adding AI dependencies to the core library.

Acceptance Criteria fulfillment

  • New @embeddedchat/ai-adapter package with IAIAdapter interface, BaseAIAdapter base class, and adapter implementations for Ollama, OpenAI, and Gemini
  • AI is entirely opt-in, passing no aiAdapter prop leaves the UI identical to today, zero footprint
  • ✨ Reply suggestions - click to get 3 AI-generated reply chips above the input
  • 📝 Chat summary - modal with a concise summary of the conversation

Video/Screenshots

2026-06-25.12-22-06.mp4

PR Test Details

Note: The PR will be ready for live testing at https://rocketchat.github.io/EmbeddedChat/pulls/pr-1334 after approval. Contributors are requested to replace <pr_number> with the actual PR number.

@Aryan-Verma-999

Aryan-Verma-999 commented Jul 22, 2026

Copy link
Copy Markdown
Contributor Author

Follow up with a few new commits:

AI Message Composer - AI-assisted message drafting directly from the composer
Thread Summarizer - Generate concise summaries for individual threads
AI Theme - Added a dedicated AI-themed UI for AI-powered features

Also attaching a short demo showcasing the new additions.

2026-07-23.01-43-59.mp4

@Spiral-Memory

Copy link
Copy Markdown
Collaborator

Hi @Aryan-Verma-999,

I went through the videos. You're moving in the right direction, but it's still not fully aligned with what we discussed.

Here are the improvements I'd like to see as early as possible:

1. Summarise Chat / Suggest Replies

  • The icons should be very subtle. The current design looks dated and doesn't feel suitable for the product.
  • Showing the summary in a popup is not an optimal experience.
  • The summary should appear as a "You only" message, meaning only the person who generated it can see it similar to how user messages looks like. Use an AI bot icon or something similar. for avatar The design is very important here, so keep it clean, subtle, and polished.
  • Instead of "Summarise", use a title like "Catch up" so it's clear that only recent messages are being summarised.
  • You can take inspiration from how Google Chat handles this with Gemini. It's implemented very well.
  • Suggested replies should be generated automatically. There shouldn't be a separate button for them.
  • For chat summarisation, keep the Catch up action at the top near the maximise icon so the experience is consistent for both channels and threads.

2. Message Composer

  • Showing a long list of options when text is selected is not the expected experience.
  • We've already discussed that the message composer should follow the same standards as the one Ayush built for Rocket.Chat last year.
  • Use a similar UI approach with subtle actions, inline text replacements, and a clean experience. It should not use an AI suggestion notification bar.
  • Please make sure you're following clean UI standards. What you've built is like a POC, but it's not yet at a product-ready quality level.

Please update the implementation based on the above feedback and let me know once it's genuinely ready for another review.

@Aryan-Verma-999

Aryan-Verma-999 commented Aug 1, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for the detailed feedback, @Spiral-Memory. I’ve reworked the AI experience to focus on a cleaner, product-ready UI:

  • Added a subtle header-level Catch up action with private “You only” summary messages.

  • Removed duplicate thread actions and the summary popups.

  • Reply suggestions now generate automatically and use the latest conversation context.

  • Rebuilt the composer around inline, streamed AI changes with accept/reject controls.

  • Reduced actions to Fix grammar, Shorten, Translate, and Emojify.

  • Tightened AI prompts for more reliable output.

It’s ready for another review.
here is the demo video for the same

2026-07-30.02-22-17.mp4

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants