Summary

Dayu Agent addresses the inefficiency of AI processing financial reports by enabling a "follow the map" approach, enhancing data confidence and traceability for investment conclusions. It serves as an assistant analyst for buy-side financial analysis, providing auditable and trackable investment reports.

How It Works

This Agent system employs an "LLM in the loop" architecture under strong host constraints, integrating with OpenClaw. It transforms financial report analysis from a broad search to a targeted "follow the map" method. Core functionalities include downloading/uploading financial reports, facilitating Q&A (single/multi-turn, WeChat), automated report writing, and rendering outputs.

Quick Start & Requirements

• Installation: Requires Python 3.11. Install via pip or offline packages. Post-installation, playwright install chromium is necessary. pandoc is optional for PDF rendering.
• Verification: dayu-cli --help, dayu-wechat --help, dayu-render --help.
• Initialization: Run dayu-cli init for interactive setup of workspace, model selection, and API keys.
• Prerequisites: Python 3.11, pip, playwright. API keys for LLM providers and optional web search tools.
• Links: GitHub Releases, CONTRIBUTING.md.

Highlighted Details

• Four Core Workflows: Financial data pipeline (download/upload), investment research Q&A (CLI/WeChat), automated buy-side analysis report writing, and rendering.
• "按图索骥" (Follow the Map) Analysis: Novel approach for AI financial report processing, improving data confidence and traceability.
• WeChat Integration: Enables Q&A via WeChat (text messages initially).

Maintenance & Community

The project provides a CONTRIBUTING.md file. No specific community channels or contributor details are highlighted.

Licensing & Compatibility

Released under the Apache License 2.0, permitting commercial use and modification with attribution requirements.

Limitations & Caveats

Key limitations include unimplemented A/HK stock report downloads, a basic FastAPI web UI skeleton, and initial WeChat text-only support. Web scraping is weak against anti-challenge measures. Features like durable memory, full FMP tool integration, earnings call audio transcription, and financial presentation extraction are pending.