Cadre documentation

Overview

Cadre is the management layer for your agent workforce. It gives a company's agents what a company gives its people: structure, delegation, oversight, and accountability. Agents hand work to each other, escalate to humans, and report everything they do.

Cadre runs as managed SaaS. Enterprise plans can deploy in your VPC or on-prem — talk to sales.

Getting started

1. Create your workspace

   Sign in to the Cadre console and connect your identity provider — Okta, Microsoft Entra, or Google Workspace. Your team logs in with SSO from day one.

2. Build your roster

   Register each agent from the console or the SDK. Every agent gets an identity, a role, and scoped credentials — you always know who is on the team.

3. Set the structure

   Decide who hands off to whom, what each agent can touch, and which decisions come to a human. Management rules, enforced automatically.

4. Put them to work

   Assign an objective. Cadre routes the work between agents, resolves conflicts, and reports back — every step recorded.

Your roster

Every agent on the roster has an identity, a role, and scoped credentials. Register agents from the console, or programmatically wherever they run:

import { Cadre } from '@cadre/sdk';

const cadre = new Cadre({ apiKey: process.env.CADRE_API_KEY });

await cadre.agents.register({
  id: 'agent://finance',
  runtime: 'anthropic:claude-opus-4-8',
  scopes: ['ledger:read'],
});

Onboarding an agent takes minutes. Offboarding takes one click — its credentials die everywhere, instantly.

Objectives & delegation

Instead of prompting four agents separately, assign one objective. Cadre routes tasks between the lead and its collaborators, resolves conflicts, and reports progress in the console:

await cadre.objectives.create({
  goal: 'Ship the Q3 pricing update',
  lead: 'agent://marketing',
  collaborators: ['agent://legal', 'agent://finance', 'agent://coding'],
  escalate_to: 'you@yourco.com',
});

When a decision is above an agent's pay grade, it escalates — to another agent with the right scope, or to a human.

Approvals & oversight

Some calls belong to people. Mark the actions that need sign-off — production deploys, payments, contracts — and approvers get the request in Slack, Microsoft Teams, or email, with one-click approve/deny.

Check a decision from your own systems via the REST API:

curl https://api.cadrehq.io/v1/actions/check \
  -H "Authorization: Bearer $CADRE_API_KEY" \
  -d '{ "agent": "agent://coding", "action": "deploy:prod" }'

{ "decision": "escalate", "to": "approvers", "reason": "deploy:prod requires sign-off" }

The record

Everything every agent does is attributed, signed, and kept — who did what, on whose instruction, and what it touched. Review it in the console, or stream it to Splunk, Datadog, or any SIEM.

Retention follows your plan: 30 days on Team, 1 year on Business, custom on Enterprise. Cadre is SOC 2 Type II audited — reports fromsecurity@cadrehq.io.

Admin

SAML SSO and SCIM provisioning keep workspace access in sync with your directory. Human access is role-based: