The Risk Network for AI

Measure and reduce AI risk at runtime with the Humanos API, build a verified risk reputation, and use it to unlock credit, insurance, yield and more.

​

Overview

Humanos is the Risk Network for AI. AI risk changes every time an agent acts, so Humanos measures it continuously, reduces it with runtime controls, and makes it provable with signed, verifiable evidence.

The Humanos API is how your systems take part in that network:

• Control risk. Verify every agent tool call against your organization’s policy before it runs, and against a mandate a person grants where the policy asks for one. See Agent Governance.
• Human approvals. Ask people to fill forms, sign documents, or approve data, and receive each approval as a signed credential. See Human Approvals.
• Measure risk. Every approval, verification, and revocation is recorded as a signed event on the credential’s chain: runtime evidence about each mandate and the agent acting under it.

An agent’s risk record is built once and can be shared with any partner across the network. Insurers, lenders, and auditors you share it with assess the agent from its verified record instead of starting from zero, which is what unlocks credit, insurance, yield, and other financial products.

---

​

API

The REST API, signed with your organization’s API key (see Authentication). Call it over HTTP, or through an official SDK: each wraps every endpoint of the same spec in idiomatic bindings for your stack, with auto-signing and webhook decryption.

Base URL: https://api.humanos.tech

TypeScript

npm install humanos

For the full SDK documentation click here.

Python

pip install humanos

For the full SDK documentation click here.

C#

dotnet add package Humanos

For the full SDK documentation click here.

Browse the endpoints →

---

​

Agent SDK

Put a TypeScript agent under your organization’s policy: every tool call checked by Humanos before it runs. It works with Claude Agent SDK, Vercel AI SDK, OpenRouter, LangChain and LangGraph, Cloudflare Agents, or an agent of your own.

npm install @humanos-ai/agent-sdk
npx humanos init

Agent SDK docs →

---

​

MCP

Connect an MCP client, such as Claude Code, Claude Desktop or Cursor, and it becomes an agent under your organization’s policy. It signs in with OAuth, and Humanos holds its key.

Endpoint: https://mcp.humanos.tech/mcp

MCP Server docs →

---

​

Concepts

Concept	Description
Agent	An AI system your organization runs, named by a DID and the key it holds. With the Agent SDK the agent holds its key; over MCP, Humanos holds it.
Tool	Something an agent can call, declared to Humanos as its server describes it and grouped by service. Humanos assesses each tool’s risk, a level from 1 to 10.
Policy	Your organization’s decision on each tool: Allow, Restrict, Delegate or Deny, with a default for the rest. You publish it from the dashboard, and Humanos checks it on every call.
Control	A set of rules over a call’s inputs. A tool set to Restrict runs within its control’s rules; a mandate control also declares the limits a person sets when granting a mandate.
Mandate	A signed credential a person grants an agent for a tool set to Delegate: under a mandate control, with the limits they set, or as an authorization in plain words.
Decision	Humanos’s answer to one call: allow or deny, with the reason. It is signed and recorded before the tool runs.
Audit trail	Every decision, outcome and mandate event, signed and hash-chained to the agent’s record.
Risk score	The agent’s risk, computed from its audit trail after each decision. It informs; it never decides a call.

---

​

Lifecycle

An agent is set up once:

1. Connect. The agent registers its key and gets its DID: npx humanos init with the Agent SDK, or onboard over MCP.
2. Declare its tools. Humanos keeps each tool as its server describes it and assesses its risk. The level orders the Policy page, riskiest first; it does not decide calls.
3. Set the policy. Choose an access for each tool, and publish the policy.

From then on, every call goes through the same check. For an agent connected, its tools declared and assessed, and its policy published, one call runs like this:

Over MCP the client calls verify and report_outcome itself, and Humanos signs for the agent, so there is no challenge. When a tool is set to Delegate and no mandate covers the call, the deny is mandate_required: the agent asks a person for a mandate once, and the next call presents it. Walk each step in Agent Governance.

---

​

Audit trail

Every consequential event in a mandate’s lifecycle, whether issuance, approval, verification, or revocation, is persisted as an immutable, cryptographically signed record chained to the credential. Get Chain with audit=true re-verifies that chain and returns a tamper-evidence report (credentials issued before the v0.2 format have no chain). You get a complete, tamper-evident history without logging anything yourself.

---

​

W3C standards

Every entity in the Humanos ecosystem, whether an automated system, an organization, or an end user, is identified by a Decentralized Identifier, or DID, that complies with the W3C DID standard. All authorizations are encapsulated as W3C Verifiable Credentials 2.0, giving each authorization a cryptographic proof, a tamper-evident structure, and a complete audit trail.

DID 1.1

W3C standard for verifiable identities with no central registry.

VC 2.0

W3C data model for the credentials every authorization uses.