Longbridge Data Connector and background responses

LongbridgeAI

Improvements

• Artifact panel and mobile menus: Resize the LongbridgeAI artifact panel or view it full screen to read research more comfortably. Some mobile “⋮” menus now open as bottom sheets for easier tapping.
• Connector management: LongbridgeAI connector management is now centralized in Settings → Connectors. The previous entries in the input field and menu have been removed. Your selected connectors are unaffected, with no reconfiguration needed.
• Memory system update: LongbridgeAI now uses a single updated memory system.

My Agents and workflow building

New features

• New models: The model list now includes qwen3.8-flash and gemini-3.8-flash, giving you more options for your Agents.
• Plugin downloads: Use the download button on the plugin configuration page to export a plugin package locally, making it easier to keep and manage copies.

Bug fixes

• Tool result retrieval: Fixed a loop in which retrieved tool results were truncated again, triggering repeated retrieval by the Agent. This reduces redundant reads within the same response.
• Loop node configuration: Fixed false incomplete-configuration warnings caused by missing variable type definitions in workflow loop nodes, reducing misleading alerts during setup.
• Workflow editing: Fixed a page crash caused by rich-text tags when a workflow node had no title, preventing this issue from interrupting editing.

LongbridgeAI

New features

• Guest Agent trials: Try public Agents in the gallery without signing in. Trial limits apply separately to each Agent, helping you find an assistant that fits your needs before choosing one.

Improvements

• Background responses and completion notifications: You can step away while waiting for a LongbridgeAI answer, and the AI will continue responding in the background. Return later or after receiving a completion message to read the answer, without having to stay in the conversation.
• AI landing page loading: The website’s /ai page now uses fewer resources for its initial view and loads content on demand, reducing the amount of data downloaded on your first visit so you can get to your research conversation sooner.
• Small-screen input field: On small screens, the input field automatically collapses into a compact pill shape to leave room for controls, making it easier to enter research questions.

Bug fixes

• Connector authorization: Fixed mismatches between the identity used to authorize a connector and the identity used at runtime, as well as repeated retries or unnecessary reauthorization prompts after authentication became invalid, reducing interruptions during research.
• Region detection: Fixed an issue that incorrectly identified users in Singapore as being in Hong Kong.

Usage changes

• Asset tool verification: When calling tools that involve assets, you must verify your trading password before proceeding.

My Agents and workflow building

New features

• GPT-6 model support: Model integrations now include gpt-6-astra through direct connections, Azure, and OpenRouter, so you can choose the route that fits your existing setup.
• Market ranking tools: Agent Platform’s built-in MCP tools now include market rankings for pre-market, regular trading, and after-hours sessions, providing data from different trading sessions for generating market updates.

Improvements

• Intelligent question classification: In customer support question classification, an LLM provides fallback classification for questions that keyword and pattern rules do not match, while maintaining backward compatibility.
• Direct model requests: Direct model calls now automatically retry when a connection is no longer valid, reducing the impact of occasional request failures on Agent runs.

Bug fixes

• Environment variable secret editing: Fixed an issue where masked values shown in the editor could corrupt configured secrets when environment variables were edited and saved.
• Workflow import: Importing a workflow no longer clears the model configuration in thinking-level nodes, reducing the need to reconfigure them after import.

Usage changes

• Workflow request access: External access from workflow HTTP request nodes and built-in request tools is now restricted to permitted destinations. If your workflow relies on external services, check that those services remain accessible.
• Chat HTML rendering: Chat messages now render only HTML tags on the allowlist; custom tags are no longer rendered. If your Agent’s output relies on custom tags, check how it displays and adjust the output accordingly.

LongbridgeAI

New features

• Longbridge Data Connector: Data access has been upgraded to the official Longbridge connector, expanding coverage to include options and other data types. Access more comprehensive, high-quality data directly from Longbridge to inform your investment analysis and decisions.
• File artifact management: In a LongbridgeAI conversation, ask AI to deliver its work as files. Files are stored independently of individual conversations for long-term access through the sidebar’s artifact management center. Use them as a record of your investment research to revisit the original reasoning and risks as you track companies over time or review past decisions, and reassess your views as new information emerges.

Improvements

• Stock quote cards: Switch time ranges and candlestick views in stock quote cards, and share market snapshots to discuss price trends and investment views with others using the same chart.
• Conversation waiting prompt: When AI needs your input in a conversation, it displays “Waiting for your reply” and sends a notification, prompting you to add research questions or relevant information and continue the analysis.
• Accessibility: Improved keyboard navigation, screen reader support and narrow-screen layouts make it easier to read analysis and ask research questions using a keyboard, screen reader or small-screen device.

My Agents and workflow building

New features

• Custom context compaction: Configure compaction prompts in Agent nodes to specify which investment objectives, risk constraints and key assumptions to prioritize retaining in long conversations. Tailor context retention to your research tasks. Context compaction is enabled by default.
• Workflow thinking level: In the workflow start node, use the new sys.thinking_level variable to reference the thinking level passed from the frontend when configuring subsequent steps.
• Variable configuration: In the configuration interface, set variables for Commands, Plugins, Hooks and OpenAPI reference channels to meet the needs of different tasks.

Improvements

• Plugin import: Import plugins with the same drag-and-drop interaction used for skills to add the extensions your Agent needs.
• Agent creation flow: The new Agent creation flow is now available to all users and replaces the previous flow.

Bug fixes

• Workflow Agent run mode: Fixed an issue that unexpectedly reset a workflow Agent’s run mode. The mode remains fixed after creation.