The Operating System for Supervised AI Trading Agents.
A safety-first platform for agent reasoning, risk review, paper trading, simulation, audit, and future gated execution. Built for teams who want intelligent agents — under human control.
- Paper
- Trading mode
- Human
- Approval required
- Audit
- Every action
- Gated
- Real execution
AI can generate trading ideas. Today's tools can't govern them.
Modern LLM agents can reason about markets — but the surrounding stack is missing. There is no shared layer for safety, governance, risk gates, audit trails, and human-in-the-loop control.
No safety layer
Agents execute without risk veto, exposure caps, or kill switches.
No audit trail
Reasoning, proposals, and approvals vanish into chat logs nobody can replay.
No human control
Either fully autonomous or fully manual — nothing in between.
A multi-agent trading OS with safety wired into the kernel.
SenpAI is the control plane between language-model agents and the markets. Reasoning, risk, approval, simulation, and audit live in one coherent runtime.
Multi-agent OS
Run many specialized agents side by side under one supervisor.
Agent cockpit
Inspect every prompt, tool call, and proposal in real time.
Risk engine
Hard gates and soft policies block unsafe actions before they ship.
Human approval
Every consequential action requires explicit operator sign-off.
Paper trading
Realistic fills against live or recorded market data — zero real capital.
Simulation lab
Replay regimes, shocks, and outages to stress-test your fleet.
Audit trail
Cryptographically signed log of every decision and override.
Real trading gate
Future supervised live execution behind a separate compliance gate.
One runtime. Composable modules.
Each module is independently deployable and shares the same identity, audit, and policy plane.
Agent Cockpit
Multi-agent control room for reasoning, proposals, and supervised actions.
Risk Engine
Pre-trade risk gates, exposure limits, and veto rules before any proposal advances.
Paper Trading Engine
Realistic simulated execution with fills, slippage models, and PnL accounting.
Simulation Lab
Replay historical regimes and stress scenarios against your agent fleet.
Robot Factory
Compose new agents from strategy primitives, prompts, and risk policies.
Marketplace
Discover, fork, and subscribe to vetted agents published by other researchers.
Real Trading Gate
Separately licensed, multi-sig gated path to supervised live execution.
What ships today
- Paper trading only — no real capital at risk
- Mock market scenarios and historical replays
- Human approval required on every proposal
- Full audit log of agent reasoning and overrides
- Investor dashboard with allocation and round status
Docker / Kubernetes for AI trading agents
- Sandbox runtime for isolated agent execution
- Strategy builder with composable primitives
- Robot marketplace for vetted, forkable agents
- External agent gateway — bring your own LLM or framework
Safety isn't a feature. It's the kernel.
Every action an agent takes passes through the same policy plane: risk veto, approval, audit. There is no bypass — even for admins.
Risk veto
Hard policy gates block any proposal that breaches exposure, concentration, or drawdown limits.
Kill switch
One operator action halts every running agent across the fleet, immediately.
Audit logs
Append-only record of prompts, tool calls, approvals, and overrides.
Permission system
Role-based scopes for investor, operator, risk, and admin surfaces.
No real trading in MVP
Real execution is disabled at the kernel level. Paper only.
Real trading gate
Future live execution requires a separate licensed, multi-sig gate.
Join the ownership reservation list.
We're inviting a small group of operators, funds, and angels into the SenpAI preview round. Reservations are provisional and subject to KYC/AML review and signed legal documentation.