Sardis primary sources — 2026-05-03

Project status in this pass

• No canonical standalone Sardis whitepaper or litepaper surfaced.
• The clearest current source of truth is the official site, docs corpus, docs-wide llms.txt index, and the policy/payments/API/MCP/protocol pages.
• Sardis is best understood as a policy-controlled payment operating system for AI agents rather than as a generic wallet SDK or one-off agent-payment demo.

Primary-source snapshots

1) Official site / top-level positioning

• URL: https://sardis.sh/
• Key excerpt:
  • The homepage says Sardis offers “Safe payments for AI agents.”
  • It frames the product around “spending rules,” “compliance guardrails,” and autonomous agent transactions.
  • It also says Sardis uses “Non-custodial MPC custody powered by Turnkey — Sardis never holds your keys,” which is central to its trust model.

2) Docs overview / system framing

• URL: https://docs.sardis.sh/docs
• Key excerpt:
  • The overview calls Sardis “the Payment OS for the Agent Economy.”
  • It says Sardis enables AI agents to make real financial transactions safely through non-custodial MPC wallets with natural-language spending policies.
  • The overview also identifies the core flow as agent request → policy engine → MPC signing → chain execution → audit trail.

3) Payments + policy engine / enforcement layer

• URLs:
  • https://docs.sardis.sh/docs/payments
  • https://docs.sardis.sh/docs/policy-engine
• Key excerpt:
  • The payments page says payments can execute via bank transfer, virtual card, or stablecoins and that every payment passes policy checks and compliance screening before chain execution.
  • The policy-engine docs show natural-language rules such as daily caps, merchant allowlists, approval thresholds, and business-hours restrictions.
  • The same docs say Sardis fails closed if policy parsing, merchant verification, compliance screening, or rule evaluation fails.

4) API + MCP / operator and agent integration surface

• URLs:
  • https://docs.sardis.sh/docs/api
  • https://docs.sardis.sh/docs/mcp-server
• Key excerpt:
  • The API reference exposes 47+ endpoints spanning wallets, payments, mandates, treasury, cards, compliance, and audit.
  • The MCP docs advertise 52 tools for Claude, Cursor, and MCP-compatible assistants.
  • Those tool groups include wallet management, payments, treasury/account linking, virtual cards, holds, checkout, and agent-to-agent messaging.

5) Protocol stack + docs index / interoperability posture

• URLs:
  • https://docs.sardis.sh/docs/protocols
  • https://docs.sardis.sh/llms.txt
• Key excerpt:
  • The protocol-stack docs say Sardis implements AP2, UCP, A2A, TAP, x402, and ACP.
  • The llms.txt index reveals a large surface spanning framework integrations, agent auth, payment objects, holds, compliance, protocol docs, and API references.
  • This matters because Sardis is clearly selling interoperability and control across the agent-commerce lifecycle, not only one settlement method.

Working conclusion

Sardis is best cataloged as an agentic payments control plane. The decisive evidence in this pass was the combination of non-custodial MPC wallets, natural-language spending-policy enforcement, mandates and approvals, treasury and card workflows, large MCP/API surfaces, and explicit protocol-level interoperability for agent commerce. It looks much more like a guardrailed payment OS that mediates between agent intent and real money movement than like a raw wallet backend or a thin MCP wrapper.