The science behind
agentic economics.
We publish our architecture papers, protocol specs, and benchmark findings openly. Good infrastructure is built on reproducible science.
Dual-State Agents: Economic + Cognitive Architecture for Autonomous Systems
We introduce the dual-state agent model — a framework where autonomous agents maintain both a cognitive state (vector memory) and an economic state (Stellar wallet). We show that economically-aware agents make measurably better resource allocation decisions under constrained budgets.
x402: A Pay-Per-Action Protocol for Autonomous Agent Economies
We present x402, an HTTP-native payment protocol that enables agents to pay for API calls, data access, and compute without API keys. Built on Stellar's fast settlement layer, x402 enables sub-second micropayments at fractions of a cent per action.
Persistent Vector Memory in Multi-Agent Systems: Compounding Knowledge Over Time
We study how persistent vector memory affects agent performance over extended operation periods. Our findings show that agents with access to team memory outperform stateless agents by 41% on complex multi-step tasks after 7 days of operation.
Soroban Guardrails: Programmable Safety Constraints for Financial Agents
We present a framework for encoding agent behavioral constraints as Soroban smart contracts on Stellar. Budget caps, spending velocity limits, and counterparty whitelists can be enforced at the protocol level, eliminating reliance on prompt-level guardrails.
Everything we build,
open by default.
Official TypeScript + Python SDK. Agent runtime, memory client, wallet integration.
TypeScriptReference implementation of the x402 pay-per-action HTTP protocol on Stellar.
TypeScriptThe default Pragmabase agent — financially-aware, memory-augmented, open-source.
PythonFailure simulation engine. Inject adversarial prompts, test hallucinations, audit decisions.
Rust