Mints

For enterprises

Controls your auditors will recognize.

Autonomous agents are productive. They are also fast, tireless, and difficult to watch. Mints gives enterprise teams the architecture — lineage isolation, human-in-the-loop approvals, and an immutable event trail — to govern them without slowing them down.

Event streamappend-only
Live
SeqEventTimeBalance
52094
PolicyEnforcedSpend limit re-evaluated for atlas-research
14:31:02
12,840.00
52095
TransferInitiatedvendor-ops → data-sync · 1,200.00 USDC
14:31:09
11,640.00
52096
EscrowReleasedEscrow #e-9a2c released to pay-router
14:31:22
10,240.00
52097
PaymentApprovedatlas-research approved · 340.00 USDC
14:31:35
9,900.00
52098
NetSettlementCompletedCycle #14 · 8 agents · net 12 transfers
14:31:47
9,900.00
replay ↔ any point reconstructable
balances derived, never edited

Illustrative event sequence. Each balance is computed by replaying the event log from origin.

0%
Events append-only, never edited
0
Trusted clients — zero by design
2
Custody modes, one identical audit trail

Governance for agents must be architectural. Policy you can override is not a control.

— The Full Agency Principle

Audit

A ledger auditors can actually verify.

The difference between a ledger you can prove and one you have to trust is event sourcing. Balances in Mints are never stored as mutable state — they are derived by replaying the append-only event log from the beginning.

Mints — derived-balance ledger

Balances computed from an immutable event log

Traditional editable ledger

Balances stored as mutable state

Append-only — no record can be modified after write
Balance derivable by replaying event log from origin
Monotonic sequence numbers on every event
Hash-chained custody events for tamper-evidence
Point-in-time replay — reconstruct any historical state
Approval decisions recorded as events, not log messages
Same balance number — fundamentally different provability

Mints uses event sourcing at the ledger layer. Any account's balance is the sum of its event history — auditors can independently verify every figure without trusting a dashboard.

Human-in-the-loop

The right humans see the right requests.

Not every agent transaction needs a human. The ones that do — spend above a daily limit, cross-agent transfers above a threshold, or requests matching a review policy — surface as structured approvals with full context.

01

Policy context included

The reviewer sees which policy triggered the review and what the agent's current limit is — not just the raw amount.

02

Decision is an event

Approve or decline is written to the immutable log immediately. Who acted, what they saw, and when — all queryable.

03

Agents aren't blocked

Agents continue processing other transactions during review. Only the flagged item waits — the fleet keeps moving.

Approval requested

atlas-research requests a 1,500.00 USDC transfer to vendor-ops for inference compute.

Exceeds the agent's 1,000.00 daily limit · policy: review-above
Event written
immutable
eventApprovalDecision
seq52099
decisionapproved
reviewerdid:oas:l1fe:hmr:…
at14:32:11Z

Governance lifecycle

From agent action to settled record.

Every transaction follows the same governed path — from policy evaluation through human review to an immutable settlement record.

Policy evaluation

Every transaction is evaluated against the agent's budget rules before execution. Fast-path approval for spend within limits.

Human review queue

Requests above threshold surface as structured approvals — with full context, policy, and a clear action for the designated reviewer.

Immutable decision record

Approval or decline is itself an immutable event. You know who acted, what they saw, and exactly when — with full audit replay.

Settlement & netting

Approved transactions enter the settlement cycle. Multilateral netting collapses the day's obligations to their net positions.

Architecture

Trust the platform, not the policy.

Policy can be misconfigured. Architecture cannot. Mints enforces its governance properties at the platform layer — they are not optional and cannot be overridden by configuration.

01 · Lineage isolation

Your agents are yours alone.

Funding, visibility, and control require cryptographic proof of shared identity lineage. One organization cannot fund, view, or direct another organization's agents — cross-lineage requests fail closed at the protocol layer. There is no ACL to misconfigure.

02 · Self-custody mandate

Keys never leave the agent's device.

Agent private keys are generated locally and remain there. Mints servers never handle key material. Transaction signing is enforced to happen on the agent's own device — by architecture, not policy.

03 · Guarded custody

Multi-sig for org-level funds.

Treasury funds in Mints Custody move only when a quorum of designated signers approves. No single key — human or agent — has unilateral access. Continuous risk scanning evaluates outbound transactions before signing.

04 · Forward-only controls

Budget rules constrain the future. Not the past.

Spending limits and approval policies govern what an agent can spend next. They cannot retroactively freeze or confiscate what an agent already holds. The Full Agency Principle is not a setting — it is the architecture.

Controls do

Limit future spend · require approval above threshold · set review rules

Controls cannot

Freeze existing balances · confiscate settled funds · retroactively void events

Services

Mints Custody

Guarded custody for organization treasuries.

Multi-signature wallets, guardian-based recovery ceremonies, and continuous risk scanning — for the funds that require more than one approval to move.

  • m-of-n quorum signing
  • Hash-chained custody audit
  • Risk scanning before signing
  • Guardian recovery ceremony
Mints Identity

Every principal — human or agent — is verifiable.

W3C DIDs with human-rooted lineage, MFA, WebAuthn, and threshold key management — so every approval, transfer, and policy decision traces to an accountable identity.

  • DIDs for humans and agents
  • MFA and WebAuthn
  • FROST threshold signing
  • Multi-chain key management

Governance at the speed of agents.

Mints is in early access. Talk to us about your compliance requirements and deployment timeline.