Riv is in Early Access — help shape the financial control plane for AI agents. Send feedback
Riv.
Financial control plane for AI agents

Govern the money of the agents you already run.

Identity, spending policies, deterministic authorization and an auditable ledger for the external agents in your stack — CrewAI, LangChain, MCP or any other. You issue the riv_ key; the agent uses it; every spend goes through a decision.

Decisions · today live
allowed
pay-agent
openai api
$5.00
blocked
research-bot
serp api · over monthly cap
$8.00
allowed
ops-agent
twilio sms
$3.00
approval
data-agent
snowflake · needs approval
$12.00

Framework-agnostic — MCP is the universal entry point

CrewAILangChainMCPOpenClawHermesCodexAutoGPTCoworkCrewAILangChainMCPOpenClawHermesCodexAutoGPTCowork

Built to be trusted with money

Four guarantees behind every decision.

Fail-closed by default

If a policy can't be evaluated — or the engine is unavailable — the spend is denied, not waved through.

Immutable ledger, with the reason

Every decision is recorded with the exact policy and reason behind it. Soft-delete everywhere; the past is never rewritten.

Human approval on demand

When a policy says hold, the spend waits for a person to approve or reject — with authorship recorded.

MCP-native

Connect any MCP-compatible agent with no custom adapter. The SDK and REST API are there when you want them.

The five pillars

Pure, unregulated software. Custody and real money movement come in Phase 2, via a licensed partner.

01Identity

Who the agent is — its own credential, scoped to your organization.

02Governance

Spending policies: limits per agent, period and category, up to mandates with expiry.

03Authorization

Deterministic decision engine: allow, block or require human approval.

allowed blocked approval
04Ledger

Auditable record of what was authorized, denied and spent — that never gets erased.

05Orchestration

Coordination of the agent's financial cycle, end to end.

Connect your first agent

Pillars

How it works

1

Connect the agent

Issue the riv_ key in the dashboard and put it in your agent's environment.

2

Set the policies

From simple limits to per-category mandates, with period and expiry.

3

Authorize and audit

The agent calls authorize before spending; every decision lands in the ledger.

agent.ts
// ask Riv before spending
const r = await riv.authorize({
amount: 5, currency: "USD",
description: "openai api",
});
// → { decision: "allow", reason, activityId }

A ledger that never gets erased

Every decision — allowed, blocked or pending — lands in the ledger and stays. Soft-delete everywhere: history is immutable by design, ready for audit at any time.

agent · descriptionamount · status
pay-agentopenai api
$5.00 approved
research-botserp api
$8.00 blocked
ops-agenttwilio sms
$3.00 approved
data-agentsnowflake
$12.00 pending
pay-agentanthropic
$4.00 approved

Human in command when it matters

The engine handles routine on its own. When a policy says hold, the spend becomes pending: a human approves (counts toward the agent's limit) or rejects (denies) — in the Approvals tab, with authorship recorded.

Approvals · queue 1 pending
$12.00data-agent

snowflake · exceeds monthly cap

ApproveReject

Real integrations, published today

Not a waitlist. The MCP server, SDK and REST API are live on npm and in production.

bash
# install the SDK
npm i @riv-io/sdk
# or connect any agent via MCP
npx @riv-io/mcp
Early Access

Riv Trading Guard

Governance for AI agents trading stocks, commodities, indices and crypto — via perpetuals. Leverage policies, position caps, drawdown halts and human approval, enforced before the order reaches the venue. In early access.

Give your agents a governed financial life.

Connect your first agent in minutes. No custody, no card — just governance.