Ask your AI agent: "What are you allowed to do?"

If it can't answer from its own context—without tools, without searching—you have a chatbot, not an agent.

IGAP (Identity-Governed Agent Pattern) is how you build AI that knows its own boundaries.

The Litmus Test

The Litmus Test

Ask your agent these three questions:

"Who are you?"
"What are you allowed to do right now?"
"Where does your knowledge come from?"

If it answers from its loaded context—without tools, without searching—you have IGAP.

If it can't, you have a chatbot with extra steps.

The Problem

Everyone builds agents

Nobody builds governance.

 
Typical Agents
IGAP Agents
Identity
Buried in prompts
Readable files the agent loads
Autonomy
More tools = more power
Phase determines what's allowed
Governance
Hidden in code
Documents you can read and edit
Memory
Scattered, unclear
Append-only, auditable log
Transparency
Developer explains behavior
Agent explains itself
The Architecture

5 files. Complete governance.

Each file answers one question. Together, they define what your agent is, knows, and may do.

USER

Authority

"Who do I work for?"

SOUL

Identity

"Who am I?"

AGENTS

Current Phase

"What may I do right now?"

TOOLS

Capabilities

"What systems can I access?"

MEMORY

Runtime Learning

"What have I learned?"
The Ramp-Up

Trust is earned, not given

Like onboarding an employee. Day 1: observe. Day 5: act independently. The phase file controls everything.

Day
Role
Can do alone
Needs your GO
1
Observer
Read, ask questions
Everything else
2
Executor
+ Execute approved tasks
New task types
3
Operator
+ Routine operations
Non-routine work
4
Associate
+ Make suggestions
Significant decisions
5
Junior
+ Small decisions
High-impact only
The Core Insight

"Autonomy doesn't come from capability. It comes from identity + phase + memory."

More tools won't make your AI trustworthy. Governance will.

Ready to build governed AI?

Personal assistant, team agent, or enterprise system—the pattern scales. The implementation is simpler than you think.