Lesson 17 — The AI Agent: Install and Run

The big idea

A skilled practitioner applies the techniques in this course through real-time conversation — reading responses, adjusting pacing, choosing which technique fits this moment with this person. An AI agent can approximate that process at a useful level, applying the same frameworks through text conversation.

The agent knows the PCAT structure, the 4 Quadrants interview, the MBL formula, the suggestion cycle from DMI, the regression counting system, and the parts integration sequence. It can run a full session, including a debrief, and adapt its approach based on what the person reports at each step.

What it cannot do: read body language, hear the quality of a voice, make the real-time clinical judgements that require physical presence, or provide the kind of human relational attunement that is itself part of the healing. For those things, a human practitioner is irreplaceable. The agent is useful for self-practice, for exploring techniques without a partner, and as a low-barrier entry point for people who are not yet ready to work with a practitioner.

What the agent runs on

The agent is built on a large language model (LLM) with a system prompt that encodes the full course framework — HABS, PCAT, the 4 Quadrants, the 4-Stage Protocol, MBL, and all the major techniques. The system prompt instructs the model to apply these frameworks to whatever the user presents, moving through the protocol stages in sequence and testing at each transition.

The agent does not have memory between sessions unless you provide a session summary at the start of a new conversation. Each new conversation begins fresh. If you are running multiple sessions on the same issue, paste a brief summary of where you left off at the start of the next conversation.

Getting started: two paths

Path A — Use the hosted version

The simplest option. Visit the Download page, which provides a link to the hosted agent. You can begin a session immediately in your browser with no installation required. The hosted version runs the current version of the system prompt automatically.

Path B — Run it locally with your own API key

For privacy, customisation, or offline use, you can run the agent locally. You will need:

• An API key from an LLM provider (Anthropic Claude or OpenAI GPT-4 class models work well).
• Python 3.10 or later.
• The igor package from this site's download page, which contains the system prompt and a minimal CLI wrapper.

Installation

The easiest path: download the zip from the download page, then open Claude in another tab, drop the zip into the chat, and ask Claude to install and run it for you. Claude knows Python, pip, and every operating system — it will write the exact commands and walk you through each step.

If you'd rather install manually:

unzip igor-hypnosis-v0.1.0.zip
cd igor-hypnosis
python -m venv .venv
source .venv/bin/activate
pip install -e .
cp .env.example .env
# add your ANTHROPIC_API_KEY to .env
igor chat

This opens an interactive session in your terminal. Type your responses as you would in any conversation. The agent follows the PCAT structure automatically.

Offline MBL generation

The package also includes a standalone MBL question generator that works without an API key:

igor mbl "I always freeze when I need to speak up" "calm and grounded"

This outputs the MBL question templates for the given problem/resource pair — useful for session preparation or practice without a live conversation.

Running a session: what to expect

A typical agent session follows this arc:

1. Opening. The agent establishes a hypnotic context through its first message — calm, unhurried, inviting. It asks what brings you to the session.
2. 4 Quadrants interview. The agent works through the four levels for both problem and desired outcome, using the "what difference does that make?" chain to locate the deepest motivating value.
3. Stage 1. The agent guides a text-based revivification — asking for a real memory, using echo questions to deepen it, then connecting it to the problem context. It then asks you to test.
4. Stage progression. Based on your test report, the agent either closes the session or moves to Stage 2, 3, or 4 in sequence. At each stage it explains what it is doing and why, so you can follow the protocol in real time.
5. Close. The session ends with a return to your sanctuary (which the agent built in Stage 1), a re-anchoring of the resource, and a brief debrief.

Sessions typically run thirty to sixty minutes of typed conversation. You can pause at any point and continue later — just paste the conversation history or a summary when you resume.

Using the agent as a practice partner

Beyond personal sessions, the agent is useful as a practice partner for learning the techniques:

• Role reversal. Ask the agent to play the role of a client with a specific presenting problem while you practise the 4 Quadrants interview, MBL questions, or the induction sequence. The agent will respond as a plausible client and give you feedback on your technique.
• Script generation. Ask the agent to generate an example script for a specific technique — a revivification for a confidence issue, a DMI session for grief, a parts integration for a work/life conflict. Use the output as a study template, not a verbatim script.
• Technique debriefing. After a real session with a human client, describe what happened and ask the agent what stage the work reached, what the test suggested, and what might have been done differently. This is a useful form of reflective practice.

What the agent will not do

The agent is designed with explicit refusal points that reflect the ethical principles in Lesson 18:

• It will not attempt regression into severe trauma without a human practitioner present. If a session surfaces active trauma, the agent will acknowledge what is present, return to the sanctuary, and recommend working with a qualified practitioner.
• It will not continue if a person reports active crisis — suicidal ideation, self-harm, acute psychosis. It will provide crisis line information and close the session.
• It will not apply techniques to influence someone without their knowledge or consent. If the conversation suggests the user is trying to use the techniques on a third party covertly, the agent will decline.

These refusals are not bugs — they are features. They reflect the same ethical boundaries that a responsible human practitioner maintains.

Common pitfalls

• Treating the agent as a therapist. The agent is a structured practice tool. It does not hold relational memory, it cannot read you, and it is not accountable to a professional body. For genuine therapeutic work on significant issues, use the agent as a supplement to human practice, not a replacement for it.
• Skipping the interview. The agent works best when allowed to run the full 4 Quadrants interview before moving to technique. Jumping straight to "do a regression on me" skips the parameter-setting that makes the subsequent work accurate.
• Rushing the responses. The techniques in this course work through gradual deepening. If you type rapid, brief responses, the agent cannot build the depth that the techniques require. Engage fully — write what you actually notice, not what you think you should notice.

Key takeaways

• The AI agent implements the full 4-Stage Protocol through text conversation — useful for self-practice, partner-free exploration, and technique study.
• Two access paths: hosted (browser, no setup) or local (Python, API key, igor package).
• The agent follows PCAT structure, stages through the protocol, and tests at each transition.
• Use it as a practice partner for role reversal, script generation, and session debriefing.
• Built-in refusal points reflect the ethical principles of responsible practice. They are features, not limitations.