How ashjha0/agentic-trader works
AgenticTrader is a research-and-education trading desk for equities and FX, not a broker-connected execution system. Its main workflow gathers market evidence, runs specialized agents to produce a sized and explained decision, then audits the result; the agentic harness validates plans, gates tool calls, and invokes a critic and reporter. Separate branches cover portfolio and quant research, backtesting and evaluation, and Python/C++ quant functionality. Several modules are represented from the tree and README rather than inspected source, so their internal call links are omitted where evidence is limited.
Main entry points: harness.py Desk harness · quant Quant core · backtest.py Walk-forward backtest · agents Specialist agents
Components 23
Agentic desk
- Desk harness
agentic_trader/agentic/harness.py - Specialist agents
agentic_trader/agents - Decision state
agentic_trader/state.py - Model access
agentic_trader/llm.py
Agentic controls
- Tool registry
agentic_trader/agentic/tools.py - Tool policy
agentic_trader/agentic/policy.py - Evidence records
agentic_trader/agentic/evidence.py - Decision critic
agentic_trader/agentic/critic.py - Audited reports
agentic_trader/agentic/reporter.py - Plan validation
agentic_trader/agentic/planner.py - HTTP and MCP
agentic_trader/agentic
Market and research
- Market providers
agentic_trader/data/base.py - Desk capabilities
agentic_trader/agentic/servers.py - Runbook retrieval
agentic_trader/agentic/rag.py
Quant and portfolio
- Alpha research
agentic_trader/alpha.py - Execution algorithms
agentic_trader/algo.py - Portfolio construction
agentic_trader/portfolio.py - Quant core
agentic_trader/quant - C++ engine
cpp/include/at/backtest.hpp
Backtest and evaluation
- Walk-forward backtest
agentic_trader/backtest.py - Backtest statistics
agentic_trader/stats.py - Evaluation protocol
agentic_trader/evaluation.py
Other components
- Research user
Connections 23
- Research user → Desk harness: requests decision
README.md - Desk harness → Plan validation: validates plan
README.md - Desk harness → Tool registry: dispatches tools
agentic_trader/agentic/harness.py - Tool registry → Tool policy: checks policy
agentic_trader/agentic/tools.py - Tool registry → Evidence records: records evidence
agentic_trader/agentic/tools.py - Desk harness → Specialist agents: runs team
README.md - Specialist agents → Decision state: writes decisions
agentic_trader/agents/analysts.py - Specialist agents → Model access: requests completions
agentic_trader/agents/base.py - Desk harness → Decision critic: checks result
README.md - Desk harness → Audited reports: builds audit report
README.md - Desk capabilities → Market providers: reads market data
agentic_trader/agentic/servers.py - Desk capabilities → Runbook retrieval: retrieves guidance
agentic_trader/agentic/servers.py - Desk harness → Desk capabilities: configures capabilities
agentic_trader/agentic/harness.py - HTTP and MCP → Desk harness: serves desk runs
agentic_trader/agentic/harness.py - Walk-forward backtest → Quant core: runs simulation
agentic_trader/backtest.py - Walk-forward backtest → Specialist agents: propagates graph
agentic_trader/backtest.py - Walk-forward backtest → Market providers: loads prices
agentic_trader/backtest.py - Quant core → C++ engine: uses compiled backend
agentic_trader/quant/__init__.py - Portfolio construction → Quant core: uses quant functions
agentic_trader/portfolio.py - Execution algorithms → Quant core: uses quant functions
agentic_trader/algo.py - Alpha research → Quant core: uses quant functions
agentic_trader/alpha.py - Evaluation protocol → Walk-forward backtest: evaluates runs
agentic_trader/evaluation.py - Evaluation protocol → Backtest statistics: computes metrics
agentic_trader/evaluation.py