Chat Client Experience
The official Chatbot system and the Agent Platform together form Longbridge's AI product, and a user's usage quota is shared between the two. This article covers Chatbot-side capabilities — Agent builders only need to be aware of them: once your Agent is published, users interact with it with these capabilities in play.
Presentation & Interaction
| Capability | Description |
|---|---|
| Conversation charts | Structured investment data is presented intuitively in conversation as professional charts, making trends/comparisons obvious at a glance; charts are interactive, so users can keep asking follow-ups about the same data set |
| Stock quote / watchlist Widgets | Quote cards and stock list components in conversation; the Stock Card supports shareable screenshots |
| Chain of thought & citations | The chain of thought is shown in stages with multiple display forms; citations appear as clickable website links |
| Answer retry branches | After a retry, switch between branches to view the answers from different runs for easy comparison |
| Voice input | Ask questions out loud like chatting, getting complex long sentences across in one go |
Memory & Personalization
| Capability | Description |
|---|---|
| Long-term memory | When long-term memory is triggered, what has been remembered is shown explicitly, with a shortcut entry to manage all memories |
| Chat history retrieval | Conversations are automatically persisted and periodically summarized; when the user asks about "the format I gave you last time/before," the AI automatically retrieves past conversations to recall it — scoped to the user, effective across Agents |
| User tags | Users can proactively maintain personalization requirements; the AI retrieves tags before conversing to generate personalized replies |
| Personalized question recommendations | Recommended questions are generated from real technicals/news/community buzz for a pool of trending stocks — the hotter the stock, the faster the refresh |
| Suggested questions & follow-ups | Suggested questions support refresh for more; follow-ups are generated dynamically from the last few rounds of Q&A |
Note: "Personalized question recommendations" and "Suggested questions & follow-ups" only take effect in the official Chatbot and are unrelated to Agents you build — the guiding questions your Agent shows users are configured yourself via the Agent preset questions setting (see Settings).
Takeaways for Builders
- Structured data output pays off: output core data in structured form, and the consumer side automatically renders it as charts/Widgets — a far better experience than plain text (see Structured Output)
- Answer style: conclusions first, no filler — the same principles apply to your Agent's prompts (see Prompt Writing Techniques)
- Personalization is handled at the platform layer: user tags and history memory take effect at the platform level, so Agents don't need to build their own user profiles — and should not; see Data Usage & Privacy Restrictions