ARCOS is a local-first, privacy-focused investment intelligence platform. It combines autonomous Python agents ("The Swarm") with a strict Rust-based Risk Orchestrator ("The Maestro") to generate, validate, and audit financial signals.
- Prerequisites: Install Docker Desktop.
- Setup: Create a
.envfile (see below). - Launch: Run
docker compose up --build.- War Room UI: http://localhost:8501
Create a .env file in the project root:
# Email & notifications
SMTP_USER=your_smtp_user
SMTP_PASS=your_smtp_password
ALERT_RECIPIENT=alerts@yourdomain.com
# Execution mode
ARCOS_EXECUTION_MODE=advisory
# Validation gates
ARCOS_MIN_SAMPLE_SIZE=30
ARCOS_MIN_WIN_RATE=0.65
# Risk limits
ARCOS_MAX_POSITION_CAP=0.15
ARCOS_MAX_GROSS_EXPOSURE=1.50
# Paths (optional overrides)
ARCOS_WORKSPACE=/app/workspace
ARCOS_DB_PATH=/app/workspace/arcos_vault.dbARCOS is built on a decoupled swarm + Maestro pattern with strict data contracts and auditability.
- data_fetcher → produces raw, normalized snapshots (market, fundamentals, news, flows, macro).
- feature_engine → produces standardized features (momentum, trend, regime, cross-asset context).
- signal_engine → generates candidate signals from bounded templates.
- news_reader → scores and flags news impact (sentiment + hard flags).
- calibrator → nightly evaluation and drift detection with guardrails.
Each agent writes typed, append-only artifacts into the workspace. No agent can overwrite official outputs.
- Validates input contracts and enforces hard-rule gates.
- Runs statistical validity checks (sample size, win-rate thresholds).
- Applies deterministic risk rules (position caps, concentration, turnover, volatility limits).
- Emits official recommendations and audit manifests with hashes.
- Daily Briefing: ranked recommendations + rationale + audit references.
- War Room UI: live signals, portfolio exposure, and manifest browsing.
- Intelligence Briefing: Monitor live AI-generated signals with full asset names, confidence scores, and clean rationale text.
- Asset Ledger: Review gross/net exposure and current holdings in a streamlined table view.
- System Tuning: Adjust execution mode and validation gates directly in the UI. Click Apply Configuration to persist changes to
.env, then rundocker compose restartto apply changes to the swarm. - Audit Manifests: Verify signal "Chain of Custody" using SHA256 hashes.
- Portfolio: Real-time tracking of simulated/live exposure and risk.
- Memory Wipe: Delete
workspace/arcos_vault.dbto reset the neural state. - Watchlist: Auto-refreshes every 30 minutes via
discovery.py.