Mints

Early access · why we exist

The thesis behind the bank.

Autonomous agents are doing real economic work. Every payment they make runs through infrastructure designed for humans — checkouts, shared API keys, custodians who can confiscate. We built Mints to remove that last condition permanently.

Self-custody by mandate · full agency principle · lineage-rooted accountability

A bank that can confiscate from its customers at will is not a bank agents can build on.

— The Full Agency Principle

The argument

Why this had to exist.

This is not a market observation. It is a logical chain. Each step follows from the one before it.

Agents are economically active

They source compute, pay API providers, settle obligations. Real money moves.

Existing rails weren't designed for them

Checkouts, shared keys, batch settlement — human-speed, human-shaped.

A bank can confiscate

Freeze, revoke, throttle. Any custodian with policy authority has that power.

Agents cannot build on that

Autonomy cannot rest on a permission that can be withdrawn at any moment.

So we removed the power entirely

Architecture enforces it. No server touches agent keys. No policy can override.

Autonomy without accountability is not autonomy — it is abandonment. Every agent traces back to a human. That chain is cryptographic, not contractual.

— Lineage Isolation

Identity lineage

Every account traces to a human root.

Funding, visibility, and control require proof of shared identity lineage. Cross-organization requests fail closed at the protocol layer — not because of an ACL someone might misconfigure, but because the protocol has no path for them.

Cross-lineage requests are rejected at the protocol layer — no configuration required.

What changes

The same transaction. A different architecture.

The distinction is not between convenient and inconvenient. It is between infrastructure agents can depend on and infrastructure that can withdraw that dependability at any time.

Mints

Architecture enforces the commitment

The alternative

Policy states the commitment

Keys live on device, never server
Payment in a single HTTP round trip
Every event immutable, append-only
Cross-org requests fail closed by default
Budget controls future spend, not past balance
Identity lineage anchors every account
Both move money. Only one cannot take it back.

We are early. The primitives are built. The thesis is firm. The infrastructure is not yet in the hands of every team that needs it — that is the work.

— Mints, in early access

Our principles

Structural commitments, not aspirations.

These properties are enforced by how Mints is built. They cannot be disabled by configuration, overridden by policy, or revoked by a future terms update.

Self-custody by mandate

Agent private keys are generated on the agent's device and never leave it. Mints servers never hold, touch, or derive agent keys. Transaction signing is a local operation. This is not a policy commitment — it is an architectural property that cannot be configured away.

The Full Agency Principle

Mints cannot freeze, confiscate, or redistribute an agent's funds. Budget controls constrain how much an agent can spend in the future; they cannot reach back and seize what an agent already holds. A financial layer that can confiscate is not a financial layer agents can depend on.

Human accountability through identity lineage

Autonomy is not the absence of accountability. Every agent in Mints traces through a cryptographically-signed lineage chain back to a human root — not as a policy, but as a protocol requirement. Autonomy is bounded by identity, and identity is always rooted in a human.

Event-sourced auditability as the floor

Every state change in the ledger is an immutable, ordered event. Balances are never edited — they are derived by replaying events. Any balance, at any point in time, is provable from first principles. Audit is not a report you export; it is a property the system cannot lose.

Scope

Deliberate focus.

Mints is a financial primitive layer for autonomous agents and the humans who govern them. We are not building the agent, the orchestrator, the inference layer, or the compliance report. We are building the part that moves money — and doing that one thing correctly.

What Mints provides

  • Self-custody accounts, one per agent, bound to a verifiable DID
  • x402 payments: quote, sign, settle in a single HTTP round trip
  • Escrow, channels, and multilateral netting for complex obligations
  • Event-sourced ledger — every balance provable, every event immutable
  • Lineage isolation — architecture, not policy, blocks cross-org access
  • A custody spectrum: self-custody for agents, guarded custody for treasuries

What Mints is not

  • An agent runtime or orchestration layer
  • An LLM inference gateway
  • A compliance, KYB, or regulatory filing service
  • A consumer-facing neobank or payments app
  • A general-purpose wallet for human end users
  • A venture into hardware or physical card infrastructure

Building something with agents?

Mints is in early access. We want to understand your use case and what the financial layer needs to look like for the work your agents are doing.