VoxCall Oracle — Voice-Powered Earnings Call Trading Agent

Live production agent that listens to earnings calls in real-time, analyzes xStock impact with a Featherless finance ensemble, and executes trades via Kraken CLI.

Deployed on Streamlit Community Cloud — free, live, open-source.

---

Architecture

┌──────────────┐     ┌────────────────┐     ┌──────────────────┐
│  🎙️ Audio     │────▶│  Speechmatics   │────▶│  Speaker-labeled  │
│  Input       │     │  RT + Batch     │     │  Transcript       │
└──────────────┘     │  Diarization    │     └────────┬─────────┘
                     └────────────────┘              │
                                                     ▼
                     ┌────────────────┐     ┌──────────────────┐
                     │  🪶 Featherless  │◀────│  Ensemble Vote    │
                     │  3-Model        │     │  sentiment/action │
                     │  Finance LLMs   │     │  confidence/ticker│
                     └────────────────┘     └────────┬─────────┘
                                                     │
                                                     ▼
                     ┌────────────────┐     ┌──────────────────┐
                     │  ⚖️ Risk Gate   │────▶│  🐙 Kraken CLI    │
                     │  Confidence     │     │  xStock Order     │
                     │  Threshold      │     │  Paper / Live     │
                     └────────────────┘     └────────┬─────────┘
                                                     │
                                                     ▼
                                            ┌──────────────────┐
                                            │  📊 Dashboard     │
                                            │  PnL + Decisions  │
                                            │  Voice Readout    │
                                            └──────────────────┘

Features

• 🎙️ Speechmatics — Real-time microphone transcription + batch audio upload with speaker diarization
• 🪶 Featherless — 3-model finance ensemble (Llama-Open-Finance-8B, finance-chat, Fin-o1-8B) with majority vote
• 🐙 Kraken CLI — xStock paper + live order execution (kraken paper buy/sell / kraken order buy/sell)
• ⚖️ Risk Gate — Configurable confidence threshold blocks low-signal trades
• 📊 Live Dashboard — Streamlit with dark theme, PnL chart, decision log, model votes, voice readout
• 🔗 LangGraph — Multi-agent workflow: Analyze → Risk Gate → Execute → Explain
• ☁️ Streamlit Cloud — Free deployment with secrets management

Targets

Speechmatics · Kraken (PnL + Social) · Featherless · Agentic Workflows · Multimodal Intelligence · Enterprise Utility

Quick Start

cp .env.example .env
# fill in your API keys
pip install -r requirements.txt
streamlit run app.py

Open http://localhost:8501.

Deploy to Streamlit Community Cloud (Free)

1. Push this repo to GitHub.
2. Go to share.streamlit.io.
3. Click New app → select your repo → branch main → file app.py.
4. Under Advanced settings → Secrets, add your keys:

   SPEECHMATICS_API_KEY = "your_key"
   FEATHERLESS_API_KEY = "your_key"
   PAPER_MODE = "true"

5. Click Deploy. Your app will be live at https://your-app.streamlit.app.

Docker (Optional)

docker compose up --build

Environment Variables

Variable	Required	Description
SPEECHMATICS_API_KEY	For live audio	Speechmatics API key for RT + batch
FEATHERLESS_API_KEY	For live ensemble	Featherless API key
KRAKEN_API_KEY	For PnL audit	Kraken read-only key
KRAKEN_API_SECRET	For live trades	Kraken secret
PAPER_MODE	No (default: true)	Paper trading mode
TRADE_AMOUNT	No (default: 0.05)	Order size
MIN_TRADE_CONFIDENCE	No (default: 75)	Minimum confidence to execute
FEATHERLESS_MODELS	No	Comma-separated model IDs
FEATHERLESS_BASE_URL	No	Custom Featherless endpoint

Live Integrations

• Speechmatics: Real-time microphone path for local; batch diarization for uploaded audio in cloud.
• Featherless: OpenAI-compatible finance ensemble with configurable model IDs.
• Kraken CLI: Paper mode via kraken paper buy/sell. Live xStock mode via kraken order ... --asset-class tokenized_asset.

Kraken Validation

After deployment, run inside the container:

sh scripts/validate_kraken.sh

Submission

See SUBMISSION.md for lablab.ai copy, demo video script, tags, and social posts.

Safety

PAPER_MODE=true is the default. All trades go through the risk gate — low-confidence signals are blocked. Switch to live execution only after reviewing your Kraken CLI authentication and risk controls.

License

MIT. See LICENSE.