Outcome infrastructure for AI agents

One task in.
A verified outcome out.

Clervo finds the right capability, qualifies route and policy, controls cost, executes within the current contract, and keeps result, evidence, and receipt boundaries legible.

One platformSix capability familiesBounded cost and proof
Clervo Apex CoreA solid faceted mechanism. A red agent task enters, cyan qualification selects a route, and gold exits after verification.
Bounded taskrequest · fixture
Verified resultevidence + receipt
AItransform
Secure Sandboxexecute
Multi-chain RPCread / simulate
Predictionresolve
Crypto Intelligencequalify
Ready · no execution

ClervoRouter

The operating layer between intent and outcome.

The locked operating model stays constant while current public availability remains a separate observed fact.

Product contract

Find. Qualify. Execute. Prove.

ClervoRouter routes bounded tasks across six permanent capability families and refuses paths that cannot satisfy the current contract.

See observed operations →
01
Read the task

Extract the intended outcome, constraints, risk, latency, and evidence requirement.

Intent
02
Find a capability

Search by task rather than by vendor or integration trivia.

Catalog
03
Qualify the route

Check current lifecycle, policy, price boundary, network/asset context, and proof requirements.

Qualify
04
Execute within bounds

Run only an eligible operation; refuse or pause when the contract breaks.

Act
05
Verify and prove

Keep normalized result, evidence, receipt, and reconciliation state separate.

Proof

Qualification and result contract

What Clervo checks—and what comes back.

This structure is the operating contract. Current availability, prices, and settlement readiness remain registry/runtime facts rather than design claims.

Before execution

Qualification contract

The agent should understand why a route is eligible before meaningful action occurs.

Capability and operationExact identity
LifecycleCurrent registry state
Policy and riskAllowed / refused
Price and approvalObserved or not bound
Network and assetExplicit when observed
After execution

Result contract

A useful result stays separate from evidence, payment state, and replay decisions.

Normalized resultTask-shaped output
EvidenceVerification material
ReceiptOnly when produced
ReconciliationVerified / refused / unresolved
Replay safetyExplicitly allowed or blocked

Interfaces

One contract, multiple interface surfaces.

These cards preserve the locked information architecture. They do not assert unsupported client compatibility or package publication.

/skill.md

Interface surface

Availability is bound elsewhere and is not inferred from this design surface.

MCP

Interface surface

Availability is bound elsewhere and is not inferred from this design surface.

TypeScript

Interface surface

Availability is bound elsewhere and is not inferred from this design surface.

Python

Interface surface

Availability is bound elsewhere and is not inferred from this design surface.

HTTP / OpenAPI

Interface surface

Availability is bound elsewhere and is not inferred from this design surface.

Trust boundary

Refusal and uncertainty remain visible.

Gold is earned only after verification. Refused and unresolved states never borrow proof color.

Verified

The required checks resolved and proof material exists for that outcome.

Refused

Policy, lifecycle, availability, cost, or approval prevents execution. No verified-gold state appears.

Unresolved

The result or settlement cannot yet be proven. Retry remains blocked until reconciliation is safe.

Operate from current truth

Find the exact operation Clervo currently exposes.

Search canonical observed routes by task, family, lifecycle, proof, and capability without promoting design fixtures into live claims.

Explore the catalog