Ai

Daily Stock Analysis v3.13.0 Expands Data Sources and Streaming AI Workflows

Daily Stock Analysis v3.13.0 Expands Data Sources and Streaming AI Workflows

daily_stock_analysis v3.13.0 is a meaningful release centered on better market data coverage, more flexible AI generation, and stronger runtime stability. The biggest changes in this version are the new Longbridge OpenAPI integration for U.S. and Hong Kong market data, Anspire Search support for real-time search and news retrieval, and streaming LiteLLM output in the standard analysis pipeline. Together, these updates make the platform more reliable for live market workflows while improving responsiveness and multilingual reporting.

What Changed

This release puts heavy emphasis on data-source resilience. Version 3.13.0 adds Longbridge as an optional preferred source for U.S. and Hong Kong quotes, with YFinance and AkShare still available as fallbacks. It also extends Tushare support for Hong Kong daily data, fixes broken Hong Kong stock name retrieval, and corrects index open-price fallback handling in efinance. These changes matter because previous gaps in U.S. and Hong Kong data could interrupt analysis quality or leave reports incomplete.

Another major addition in v3.13.0 is Anspire Search integration. When configured, the platform can use it for real-time market and news retrieval, giving the analysis stack another semantic search path beyond existing providers. The search layer was also tuned so Chinese-language A-share news gets prioritized more reliably instead of defaulting too heavily toward English results.

On the AI workflow side, the ordinary stock-analysis path now supports LiteLLM streaming output. In practical terms, this version introduces finer-grained SSE task progress events and allows analysis responses to stream while preserving compatibility with providers that do not support streaming by automatically falling back to non-streaming calls. Web-based model configuration was also improved with a unified /v1/models discovery flow, making it easier to pull and store available model lists by channel.

Version 3.13.0 also strengthens agent and orchestration behavior. The release fixes inconsistent AGENT_MAX_STEPS handling, improves graceful degradation when skill agents fail, and adds clearer error reporting for SSE cleanup and background task failures. That makes the multi-agent and skill-aware execution path more predictable, especially in longer or more complex runs.

Database reliability received attention too. SQLite writes were reworked with atomic batch upserts, WAL mode, busy timeouts, and limited retries. For users running concurrent batch analysis jobs, this should reduce lock contention and lower the risk of write-path failures during heavier usage.

Finally, reporting and localization got polished in this version. The market-review workflow now respects REPORT_LANGUAGE=en more consistently, avoiding mixed-language output where English content was paired with Chinese headings or wrappers. The task status API also now includes current-price data more reliably, improving front-end report presentation.

Why It Matters

The significance of v3.13.0 is that it improves both the freshness of market inputs and the smoothness of AI output delivery. For an analysis product, better data-source redundancy is not just a convenience—it directly affects trust in the generated report. Adding Longbridge and improving Hong Kong market support reduces blind spots in cross-market workflows.

The streaming LiteLLM changes are especially important for user experience. Instead of waiting for a long analysis cycle to finish before seeing progress, users can now get more responsive feedback during execution. That makes the platform feel faster and more transparent, which is valuable in time-sensitive market analysis environments.

The release also shows a mature operational focus. Improvements to agent step limits, fallback behavior, SSE exception visibility, and SQLite durability indicate that the project is not only adding features, but also tightening the runtime foundations needed for production-like use.

Overall, daily_stock_analysis v3.13.0 is a strong version update because it meaningfully improves data reliability, AI responsiveness, and workflow robustness without breaking the existing interface contract. For users analyzing A-shares, Hong Kong equities, and U.S. stocks through AI-assisted pipelines, this version should deliver a more stable and capable experience.

Official Source: https://github.com/ZhuLinsen/daily_stock_analysis/releases/tag/v3.13.0