Sardis

  • Name: Sardis
  • URL: https://sardis.sh/
  • Category: AI-agent payments infrastructure / policy-controlled wallet-and-payment control plane / non-custodial MPC payment OS
  • Summary: Sardis is best understood as a policy-controlled payment operating system for AI agents rather than as a simple wallet SDK, generic AI-agent framework, or one-off MCP demo. In this pass, the clearest first-party evidence came from the homepage, docs overview, payments and policy-engine pages, the REST API reference, the MCP-server docs, the protocol-stack page, and the docs-wide llms.txt index. Those materials describe a stack built around non-custodial MPC wallets, natural-language or JSON spending policies, spending mandates, compliance screening, multi-rail payment execution, treasury funding/withdrawal, virtual cards, audit events, and a large tool surface for agent frameworks. The key distinction is that Sardis is packaging the control layer that sits between agent intent and real money movement, with explicit guardrails, approvals, and protocol interoperability rather than just exposing raw wallet access.
  • What it does:
    • Creates non-custodial MPC wallets for agents, with policy configuration and chain/token context managed through API, SDK, and MCP tools
    • Enforces natural-language or structured-JSON spending policies covering limits, merchant allowlists/blocklists, approval thresholds, time restrictions, and compliance checks
    • Executes payments across stablecoins, bank transfers, and virtual cards, with policy and sanctions screening before chain execution
    • Supports spending mandates, approval requests, and audit trails so agent payment authority can be scoped, checked, and revoked
    • Exposes treasury workflows for bank funding, bank withdrawal, external-bank-account linking, and balance snapshots
    • Offers virtual-card issuance and lifecycle management alongside holds and capture/release flows for reservation-style payments
    • Integrates with MCP and SDK surfaces for Claude, Cursor, and other agent frameworks, with dozens of tools and framework-specific guides
    • Implements agent-commerce and coordination protocols including AP2, UCP, A2A, TAP, x402, and ACP
  • Key claims:
    • The homepage brands Sardis as “Safe payments for AI agents” and says users can “set spending rules, enforce compliance guardrails, and let your agents transact autonomously,” which directly frames the product as a safety and control layer for agent spending
    • The docs overview calls Sardis “the Payment OS for the Agent Economy” and describes infrastructure enabling AI agents to transact through non-custodial MPC wallets with natural-language spending policies
    • The policy-engine docs say every transaction passes through policy checks before signing and broadcasting, with natural-language rules like per-transaction caps, merchant allowlists, business-hours restrictions, and human-approval thresholds
    • The same policy docs say the engine fails closed if policies cannot be parsed, compliance screening times out, merchants cannot be verified, or rule evaluation fails, which is a strong clue that the trust model centers on denial-by-default controls rather than permissive wallet tooling
    • The payments docs describe execution via bank transfer, virtual card, or stablecoins, and position policy check, compliance screening, MPC signing, and chain execution as one pipeline
    • The API reference exposes wallets, payments, mandates, treasury, cards, compliance, and audit endpoints, which makes Sardis look like a full payments control plane instead of a thin wallet wrapper
    • The MCP-server docs advertise 52 tools for Claude, Cursor, and MCP-compatible assistants, including wallet, payment, treasury, card, holds, checkout, and agent-to-agent operations
    • The protocol-stack docs show Sardis implementing AP2, UCP, A2A, TAP, x402, and ACP, which suggests the company is positioning itself as interoperability infrastructure for agent commerce, not just a single settlement rail
  • Whitepaper: No canonical standalone Sardis whitepaper or litepaper surfaced in this pass. The clearest current source of truth was the official site, the docs corpus and llms.txt index, the policy/payments/API/MCP/protocol pages, and the public API-doc surfaces; see ../whitepapers/sardis-primary-sources-2026-05-03.md.
  • Sources:
  • Last reviewed: 2026-05-03 UTC