| title | emoji | colorFrom | colorTo | sdk | app_file | python_version | pinned |
|---|---|---|---|---|---|---|---|
FinAI-Agent |
💰 |
indigo |
green |
streamlit |
app.py |
3.13 |
false |
Finance-Agent is your chat-native CFO co-pilot.
Ask plain-English questions about your company’s Excel ledger, and get instant answers with context + charts.
https://huggingface.co/spaces/Sbboss/FinAIAgent
You ➜ “Compare EBITDA and cash runway for Q1 vs Q2, convert EUR to USD.”
Agent ➜ “EBITDA Q1: $1.42 M, Q2: $1.65 M (+16.4 %).
Cash runway improved from 7.1 → 8.3 months.”
- Natural-Language Querying — Gemini 2.5 Flash interprets finance jargon and casual questions alike.
- Dynamic Tool Routing — LangChain React agent picks one or more of 7 custom Python tools (revenue variance, gross-margin %, OpEx breakdown…).
- Smart Currency Handling — Detects EUR rows, fetches FX sheet, normalizes to USD.
- Date Inference — “This year”, “last 3 months”, “Jun’25” → precise periods.
- Chart Factory — Generates PNGs via Matplotlib (line, bar, area, stacked).
- Streamlit UI — Slack-style sidebar, message persistence, spinner feedback.
- Excel Plug-and-Play — Works with a single
data.xlsxcontaining 4 sheets:actuals,budget,cash,fx. - Fallback PythonREPL — If no tool fits, agent writes ad-hoc Pandas code.
- Pytest Suite — Automated tests for tool selection, calc accuracy, caching, and rendering.
- One-click Deploy — Just
streamlit run app.py.
┌──────── Chat UI (Streamlit) ───────┐
│ User Input + Chat History │
│ ↓ finance_agent.invoke() │
└────────────────────────────────────┘
│
▼
┌── LangChain AgentExecutor ──────────────┐
│ Prompt (system + history + input) │
│ LLM: Gemini 2.5 Flash │
│ React tool-calling loop │
└──────────────┬──────────────────────────┘
│ selects & runs
┌──────────────┴──────────────────────────┐
│ Custom Tools │
│ • get_revenue_variance │
│ • get_gross_margin_pct │
│ • get_opex_breakdown │
│ • get_ebitda_proxy │
│ • get_cash_runway │
│ • plot_chart │
│ • code_analysis (PythonREPL) │
└─────────────────────────────────────────┘
finance-agent/
├─ agent/
│ ├─ agent.py # creates cached AgentExecutor
│ ├─ utils.py # numeric calculations
│ └─ tests/ # pytest suite
├─ data.xlsx # sample ledger
├─ app.py # Streamlit front-end
├─ requirements.txt
└─ .streamlit/
└─ secrets.toml # custom theme & font
git clone https://github.com/Sbboss/FinAIAgent.git
cd FinAIAgent
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py🏃 Quick Start
Drop your company ledger into data.xlsx with sheets: actuals, budget, cash, fx.
- Run the app:
streamlit run app.py- Ask away:
“Trend gross margin % monthly for 2024.”
“Why did revenue miss budget in Nov 25?”
“How many months of runway if burn stays flat?”
👉 Charts appear inline; numeric answers include deltas vs budget and YOY.
🔬 Testing
pytest -v
Covers: Tool parameter validation Correct tool selection via intermediate steps Currency conversion accuracy Streamlit image rendering Caching behaviour (st.cache_resource)
🛠️ Configuration
| Setting | File / Env | Default |
|---|---|---|
| Google Gemini API key | GOOGLE_API_KEY |
put in secrets.toml |
| Excel file path | data.xlsx |
— |
| LLM temperature | agent | 0.2 |
🌅 Roadmap
- 🔄 CSV and SQL connectors
- 🗃️ Vector memory for conversation continuity
- 🌐 Multi-LLM failover (Gemini → GPT-4o)
- 📈 Automatic dashboard snapshots to PDF
- 💬 Slack / Teams integration
- 🎥 Demo GIF: live chat → chart → answer
🙌 Contributing
- Fork & branch (feat/your-feature)
- Add tests (pytest -k your_test)
- Submit PR with concise description
✅ All significant changes must pass CI and keep coverage ≥ 95%.
📝 License MIT — see LICENSE Finance-Agent turns spreadsheets into strategy. Ask. Analyze. Act. ⚡