FinClaw is the first open-source, finance-specific autonomous AI agent execution framework, co-developed by the AI Finance Lab at Shanghai University of Finance and Economics. It addresses the limitations of general-purpose AI in professional financial scenarios. FinClaw introduces six specialized "Financial Lobsters" covering banking, securities, insurance, funds, futures, and trusts. Each is tailored to its respective sector, enabling AI to think and act like experienced financial professionals while staying aligned with real business operations and regulatory requirements. FinClaw features over 1,000 high-quality, reusable financial Skills, covering everything from data acquisition and analysis to result output. To ensure stable and professional data handling, FinClaw includes a unified financial data abstraction layer that leverages intelligent routing and failover mechanisms to mask underlying data source differences and maintain a consistent interface. The framework supports one-click Docker deployment, packaging the runtime environment and services together for out-of-the-box use. Natively compatible with the OpenClaw Agent OS architecture, FinClaw requires no additional API keys or complex configuration. Users can invoke financial Skills directly through conversation or the command line, orchestrating them alongside general-purpose capabilities.
FinClaw is an open-source, autonomous AI agent execution framework specifically designed for the financial industry. It is a joint effort led by Professor Liwen Zhang, Director of the Shanghai Financial Intelligent Engineering Technology Research Center and affiliated with the School of Statistics and Data Science at Shanghai University of Finance and Economics, and his AI Finance Lab (AIFinLab). Unlike general-purpose agents, FinClaw takes a vertically specialized approach, creating six distinct Financial Lobsters. This allows the AI to replicate the reasoning and behavior of experienced financial professionals, catering to the business scenarios and regulatory demands of the six core sectors: banking, securities, insurance, funds, futures, and trusts.
FinClaw provides a dedicated agent for each financial sector. Each lobster has a clear role and capabilities aligned with its specific domain, ready to solve practical business problems.
| Lobster Type | Positioning | Core Capabilities |
|---|---|---|
| Banking Lobster | Expert in Corporate Credit & Proprietary Investment | Credit approval, industry research, asset allocation, credit risk monitoring |
| Fund Lobster | Precision Tool for NAV Attribution & Performance Diagnosis | Investment research backtesting, performance attribution, portfolio compliance, FOF portfolio construction |
| Securities Lobster | Insightful Radar for Investment Research & Business Development | Investment banking due diligence, industry research, institutional roadshows, margin financing |
| Insurance Lobster | Expert in Liability Matching & Asset Allocation | Product comparison, coverage analysis, underwriting & claims, compliance & risk control |
| Trust Lobster | Specialist Architect in Non-Standard Business & Wealth Succession | Family trusts, non-standard asset valuation, ABS calculation, municipal infrastructure project risk control |
| Futures Lobster | All-Weather Trader for Market Review & Trading Analysis | Major contract review, calendar spread analysis, industry driver analysis |
In practice, the Banking Lobster can quickly generate industry credit risk monitoring lists and strategic asset allocation ideas for proprietary accounts. The Securities Lobster accurately captures policy and market trends, helping research and brokerage operations stay ahead. The Insurance Lobster provides asset allocation strategies tailored to insurers' specific needs, covering a wide range of common tasks in the insurance business. The Fund Lobster can break down why an ETF underperformed its benchmark in one click, pinpointing the source of drag on net value. The Futures Lobster can complete a review of major contracts within minutes, clarifying the reasons behind end-of-day volatility. The Trust Lobster offers professional advice for scenarios like family trusts and non-standard asset valuations, helping to formulate complete plans.
Tailored to the characteristics of the six major financial sectors, FinClaw selects and deeply adapts high-frequency business capabilities, filling the gaps left by general-purpose AI in professional financial contexts. By referencing typical financial workflows, these capabilities are standardized, abstracted, and structured into a reusable, high-quality Skills set. This set handles everything from data acquisition and processing analysis to final output. The capability system is uniformly connected to multi-source real-world data, ensuring stable, consistent, and professional task execution.
These 60 selected Skills are organized into six suites, one for each sector, with 10 Skills per suite covering core business analysis capabilities. Each Skill is backed by dedicated, professional, and accurate data sources.
FinClaw's Skills system is not just a simple sector-based breakdown. It starts from the underlying capability structure, abstracting and standardizing common functionalities across financial business. These capabilities can be flexibly combined and applied across the six major financial application scenarios. The over 1,000 Skills cover a wide range of detailed aspects within the financial business, distributed as follows:
| Category | Count | Coverage |
|---|---|---|
| Banking Operations | 155 | Corporate/Retail/Wealth Management/Risk Management/Compliance Operations |
| Investment Research Assistant | 357 | Company Research/Industry Research/Announcement Analysis/Due Diligence |
| A-Share Investment Research | 174 | Valuation/Financials/Technical/Funds/Sentiment/Macro/Stock Selection |
| Insurance Operations | 87 | Underwriting/Claims/Products/Coverage/Marketing/Compliance |
| Fund Operations | 42 | Screening/Allocation/Dollar-Cost Averaging/Attribution/Monitoring |
| Securities Operations | 20 | Brokerage/Investment Banking/Asset Management/Margin Financing |
| Trust Operations | 20 | Product Analysis/Family Trust/Post-Investment Monitoring |
| Data Sources | 60 | AkShare/TxEast Money/Hexun/Juchao/FRED/etc. |
| Risk & Compliance | 33 | Compliance Checks/Risk Alerts/Regulatory Reporting |
| General Tools | 54 | Document Processing/Frontend Design/Skill Creation |
| News & Sentiment | 8 | Financial News/Sentiment Analysis/Public Opinion Monitoring |
| Quantitative Tools | 8 | Backtesting/Factors/Portfolio Optimization/Visualization |
| Others | 7 | AI Stock Selection/Commodity Data/Atomic Tasks |
All these Skills are freely available, require no registration, and are ready to use upon installation.
FinClaw incorporates a standardized data access layer, defining a unified schema, code specifications, and access protocols across data sources. The system uses intelligent routing and failover degradation to dynamically manage sources like AkShare, TxEast Money, Hexun, and Juchao. This masks underlying differences while ensuring stable data service delivery. Upper-layer agents and Skills access data through this unified interface.
The core logic for data invocation is: User Request → cn-stock-data (Unified Entry) → Intelligent Routing → efinance / akshare / adata / ashare / snowball. Priority routing is set for different data types, and the system standardizes code formats (e.g., SH600519) and field names (English, snake_case). Automatic fallback is supported, so the upper-layer Skill never needs to know which underlying data source is being used.
FinClaw supports both standard local installation and Docker containerized deployment. The Docker option packages the runtime environment, dependencies, and core services together for simple, out-of-the-box startup. Container isolation prevents environment conflicts, enhancing system stability and reproducibility, while also improving security at the data and business levels. This approach offers good scalability and environmental adaptability, making it suitable for enterprise-level deployment and production environments.
# 1. Install OpenClaw
curl -fsSL https://openclaw.ai/install.sh | bash
# 2. Clone FinClaw
cd ~/.openclaw/workspace
git clone https://github.com/aifinlab/FinClaw .
# 3. Install dependencies
pip install akshare pandas numpy requests
# 4. Start
openclaw start
# 1. Start the OpenClaw container
docker run -d --name openclaw -p 18789:18789 openclaw/openclaw:latest
# 2. Copy FinClaw into the container
docker exec openclaw mkdir /home/node/.openclaw/workspace/
docker cp ./FinClaw openclaw:/home/node/.openclaw/workspace/FinClaw
# 3. Restart the container
docker restart openclaw
FinClaw is natively compatible with the OpenClaw Agent OS architecture, seamlessly integrating with its message routing, session management, skill registration, and permission control systems. The framework requires no extra API keys or complex environment configuration; after a standard startup, it is ready to perform financial tasks. Users can invoke financial Skills directly within a conversation or orchestrate them alongside general-purpose capabilities for collaborative execution.
After startup, simply interact with OpenClaw in the conversation interface. For example:
# List all skills
ls -1 skills/
# Find skills by category
ls -1 skills/ | grep "^bank-" # Banking operations
ls -1 skills/ | grep "^a-share-" # A-share related
ls -1 skills/ | grep "^fund-" # Fund related