Quick definition
The orchestration framework used by Treeova's AI trading agents to manage multi-step reasoning workflows, tool calls, and decision trees.
Why a graph and not a script
Trading decisions branch. "If conviction is above 70 and IV Rank is above 40, run the credit-spread sizer; otherwise run the alert-only path." Modeling this as a graph lets the runtime parallelize independent branches, retry failed nodes in isolation, and snapshot state at every step so a crash mid-run can resume rather than restart.
Persistence and replay
Every node's input and output is persisted to Treeova's Lossless Context Management ledger. This makes runs auditable (the user can see exactly what each step received and produced) and replayable (the same input set produces the same decisions, which is essential for debugging an agent that misbehaved last Tuesday).
Treeova-specific extensions
Treeova layers its own contracts on top of LangGraph: modality enforcement (alert-only chains physically cannot dispatch trade tools), per-leg budgets, model-aware healing with locked first-leg targets, and atomic continuation handoff for runs that exceed wall-clock limits. These extensions are what make LangGraph production-grade for financial workloads.