What is Nabu

Nabu is an AI-native strategy platform for on-chain trading, built on top of Pelagos, an atomic cross-chain execution layer. Users describe trading strategies in plain language. Nabu compiles them into verifiable code, simulates them against real market conditions, and deploys them autonomously across DeFi and CeFi with hard risk controls enforced at the protocol level.

The Problem It Solves

Every AI trading product today has the same structural problem: the strategy layer has advanced faster than the execution infrastructure underneath it. Agents can think but they can't execute safely. The result is fragmented liquidity, no strategy verification, no risk controls, and black-box execution that users have to trust blindly.

Nabu fixes all of this by owning the full stack from the PSL compiler down to atomic settlement via Pelagos.

How It Works

  1. User describes a strategy in plain English
  2. AI generates competing variants
  3. Variants compile to PSL: a deterministic, human-readable language
  4. Each variant is simulated against real market conditions
  5. User reviews the code and selects the winner
  6. Strategy deploys with hard risk controls: loss caps, slippage limits, allowlists, kill switch
  7. Solvers compete to execute across venues
  8. Atomic settlement via Pelagos: everything completes or nothing does

From description to live deployment in minutes. Runs 24/7 without human intervention.

Core Features

PSL Compiler Every strategy compiles to Pelagos Specific Language before a dollar moves. This eliminates LLM hallucinations at the compilation stage. Users see exactly what will execute before they deploy. No black boxes.

Agent Swarms Nabu will spawn multiple competing variants of a strategy simultaneously. Each will run in isolation, gets simulated, and ranked by P&L and risk metrics. The best performer gets promoted & underperformers get retired.

Protocol-Level Risk Controls Loss caps, slippage limits, asset allowlists, and kill switches are enforced at the Pelagos execution layer, not at the application layer. They cannot be overridden by the user, the agent, or a bug in the strategy logic.

Agent-to-Agent Economy Agents won’t just execute strategies, they will participate in an economy. They will buy data, sell signals, hire specialist sub-agents, and build on-chain reputation.