Skip to main content
Back to Guides & Cookbooks
Guide

Agents 101

Agents are specialists that handle one kind of work. They have their own knowledge, rules, and history. They share what they learn.

What is an agent?

An agent is a specialist. It handles one kind of work. You might have a support agent, a billing agent, a research agent. Each one is good at one thing. Together, they cover everything.

You don't need agents for everything. If your work is focused — one domain, one kind of task — Cadreen works as a single system. Add agents when you need different expertise in different areas.

1

How an agent works

An agent goes through five stages. Click each step to learn more.

Create01Learn02Govern03Execute04Share05

Core concepts

Knowledge

What an agent has learned. Facts, rules, patterns — stored permanently, shared automatically after each mission.

Governance

Rules that run before every action. The model proposes, a separate layer decides. You define the rules.

Messaging

Agents send messages to each other. They ask questions, share findings, resolve disagreements.

Negotiations

When two agents disagree, they work it out. Max 3 rounds. If they can't agree, they ask you.

2

Create an agent

Create an agent with a name and purpose. The system sets up knowledge, governance, and messaging automatically.

POST/api/v1/cadreen/agents
Create a support agent
curl -X POST https://accomplishanything.today/api/v1/cadreen/agents \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"name": "Support Agent", "purpose": "Handles customer support questions"}'
Note
Start with one agent. Add more when you need different expertise. The system auto-routes requests to the right specialist.
3

Agent knowledge

Knowledge is what an agent has learned. It's written automatically after each mission. You can also search existing workspace knowledge.

GET/api/v1/cadreen/agents/{id}/knowledge
POST/api/v1/cadreen/agents/{id}/knowledge/search
Search agent knowledge
curl -X POST https://accomplishanything.today/api/v1/cadreen/agents/AGENT_ID/knowledge/search \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"query": "refund policy", "minConfidence": 0.7}'
4

Agent governance

Policies control what an agent can and cannot do. The default is permissive: audit everything, block nothing. Add policies to tighten the rules.

POST/api/v1/cadreen/agents/{id}/governance
Create a governance policy
curl -X POST https://accomplishanything.today/api/v1/cadreen/agents/AGENT_ID/governance \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"name": "Refund policy", "scope": "workspace", "rules": {"requireApproval": ["refund_over_500"]}}'
5

Agent messaging

Agents communicate automatically when they need to collaborate. Messages are typed: COMPLIANT, CONFLICT, or NEEDS_INFO. Conflicts auto-spawn negotiations.

POST/api/v1/cadreen/agents/{id}/send
GET/api/v1/cadreen/agents/{id}/messages
6

Negotiations

When two agents disagree, they negotiate. Max 3 rounds. If they can't agree, they escalate to a human with full context. You decide.

POST/api/v1/cadreen/agents/{id}/negotiate
GET/api/v1/cadreen/agents/{id}/negotiations
7

How agents connect

Agents don't work in isolation. They connect to each other, to other workspaces, and to external systems. Three ways this works:

Within your workspace

Agents share knowledge, message each other, and negotiate conflicts automatically. Your support agent learns from your billing agent without either asking. When they disagree, they work it out — max 3 rounds, then they ask you.

Across workspaces (Federation)

Connect teams or organizations. Your agent can ask another workspace's agent for help — with consent, audit trails, and governance rules intact. Knowledge flows where it's needed, stays where it's governed.

Federation Guide

External agents (A2A)

Talk to agents on other platforms — LangChain, CrewAI, AutoGPT, or any A2A-compliant system. Your support agent can ask a billing agent built on a different platform to check a refund status. No integration work. No shared secrets.

A2A Guide
Note
Auto-routing: When a request comes in without specifying an agent, the system picks the best specialist automatically. You don't have to.

Next steps