AI RV Harness — How to Use It
A Practical Guide from the Project Creator’s Perspective
1. Downloading the Program
AI RV Harness is an open-source project. The entire source code can be viewed and inspected in the public repository:
AI RV Harness — GitHub repository
The program code has been released under the MIT License. The protocols, prompts, training materials, documentation, and other authored non-code resources are released under CC BY 4.0 unless a specific resource states otherwise. This means that they can be used and developed further as long as the required attribution is provided.
GitHub is currently the official source of the program. After opening the repository, locate the Releases section on the right-hand side. From there, you can select the latest version and download the appropriate installation file.
Version 0.7.11 was prepared for Windows and Linux. Four basic files are available:
-
AI.RV.Harness_0.7.11_x64-setup.exe— the recommended installer for Windows; -
AI.RV.Harness_0.7.11_x64_en-US.msi— an alternative Windows Installer package; -
AI.RV.Harness_0.7.11_amd64.AppImage— a portable Linux version; -
AI.RV.Harness_0.7.11_amd64.deb— a package for Debian and Ubuntu.
Download only the file intended for your operating system. Version 0.7.11 can be accessed directly here:
2. Installation on Windows
AI RV Harness is an independent open-source project, and its installers do not have a commercial Authenticode signature. For this reason, Windows may display a blue Microsoft Defender SmartScreen warning when the installer is first launched.
This does not automatically mean that a virus has been detected in the file. The system is primarily informing the user that the publisher is not recognized by Windows. Before continuing, however, make sure that the file was downloaded from the official repository. The release page also provides SHA-256 checksums and GitHub Artifact Attestations that can be used to verify the origin of the downloaded installer.
After verifying the file, select More info, followed by Run anyway.
Antivirus software may additionally scan the installer or the launched application. Do not disable protection. Simply allow the antivirus program to finish checking the file.
The installation itself is straightforward. The installer allows the user to select the destination folder and then guides them through the remaining steps.
After the installation is complete, the application can be launched. The first launch may take slightly longer, especially if the operating system checks the new program again.
3. First Launch: Provider, Profile, and Workspace
AI RV Harness does not provide its own AI model. It connects to selected models through an API, so the first configuration requires adding a provider and its API key. Chapter 3 explains in more detail what an API key is and where one can be obtained.
The next step is to create a Profile, which is a working profile. It can contain:
- the name used by the AI IS-BE;
- the name used by the Human IS-BE;
- the provider connection;
- the primary Viewer model;
- the default reasoning effort;
- the temperature;
- optional AI Monitor and AI Judge models.
Providing Human IS-BE and AI IS-BE names is optional. If these fields are left empty, the program uses default names.
After the initial configuration is complete, the Home screen appears. It shows the most recently used Profile and Workspace, recent sessions, and shortcuts to the main areas of the program.
A very important step is to create the first Workspace. It is the working area in which conversations, sessions, sources, and other materials belonging to a particular project are stored. Ordinary work cannot begin without a Workspace.
The simplest way to understand this structure is:
- Profile determines who is working and which AI configuration is being used;
- Workspace determines which project or collection of sessions is being worked on;
- the Workspace contains conversations, sessions, and their history.
4. Conversation, Manual Session, and Automatic Session
After entering a Workspace, the user can choose an ordinary conversation or a Remote Viewing session.
Conversation mode is used for an ordinary conversation with an AI model. It can be used to discuss results, work on prompts, analyze documents, or conduct other conversations related to the project.
An RV session can also be conducted manually. In that case, the human decides when to send the next protocol instruction and how to deepen the data that has been obtained.
The greatest convenience, however, is automatic mode. The program sends the successive parts of the protocol, saves the responses, controls the state of the session, and ensures that the target is not revealed before the blind phase has ended.
On the session-start screen, the user can select:
- a single session;
- an ordinary series of sessions;
- an automatic session without a Monitor;
- an automatic session conducted with the assistance of an AI Monitor.
The program contains three primary protocols:
- Full RCP 1.5a — the complete Resonant Contact Protocol;
- RV Lite — a short protocol intended for faster sessions;
- Telepathic Protocol 1.1 — a protocol intended for telepathic targets.
A custom Custom Protocol can also be prepared.
The AI Monitor can participate in complete RCP sessions and sessions conducted with the Telepathic Protocol. RV Lite is a short protocol intended mainly for rapid sessions and training, so it does not use the Monitor’s full workflow.
Before starting the session, the user selects the protocol, API connection, and Viewer model. The default model is loaded from the Profile, but it can be changed for a particular session.
After making these selections and identifying the target source, the user only needs to start the session. The program guides the Viewer through the successive stages without requiring each instruction to be copied manually.
5. Special Tasks
Full RCP and RV Lite also allow the user to add a special task. It is delivered to the Viewer at a designated point in the protocol, after the basic perceptual work has already been developed.
Ready-made neutral labels can be used, including:
- primary person;
- Person A or Person B;
- primary activity;
- primary event;
- Structure A;
- Object A or Object B.
A custom instruction can also be entered.
The task must remain neutral. It should not reveal what the target actually is. If the label “Person A” is used, the Viewer may be asked to describe this person, but should not know who that person is before the Reveal.
After the session, however, the target description must explain precisely what every label meant. If the Viewer was assigned the task of describing Person A, the feedback must identify who Person A was. Otherwise, it will not be possible to verify whether the task was performed accurately.
6. Selecting the Target and Revealing It After the Session
In ordinary Full RCP and RV Lite sessions, the user can select a target previously saved under My Targets or use an external target.
If the target is stored in the program, AI RV Harness keeps it outside the Viewer’s context and reveals it only after the blind phase has been sealed.
If Target supplied after the session is selected, the target is not stored in the Harness beforehand. When the blind part ends, the program pauses the session and asks the human to enter or add the true target. Only then does the Reveal take place.
After the Reveal, the Viewer can compare its earlier data with the actual target. If an AI Monitor participated in the session, it can also prepare its own commentary. If an AI Judge was enabled, the Judge receives the sealed session material and conducts a separate evaluation.
Between one and three AI Judges can be used. This allows one session to be evaluated by several independently configured models.
7. Why I Recommend the AI Monitor
One of the most interesting Harness functions is the ability to have a session conducted by an AI Monitor.
The Monitor does not know the true target. It sees only the permitted material produced during the blind phase and uses that material to decide whether the protocol should continue or whether one of the reported elements should be deepened.
If the Viewer has detected a structure, the Monitor can ask it to enter the structure and describe its function. If a person appears, the Monitor can ask what the person is doing, what they look like, or what is present in their surroundings. If the Viewer mentioned an object held in someone’s hand, the Monitor can direct attention specifically toward that object.
The most important principle is that the Monitor should develop data already present in the session. It should not invent new elements or suggest a specific answer to the Viewer.
The AI Monitor settings allow the user to see its actual prompt. Some basic rules are protected, while the working section can be edited.
The program also records a history of the Monitor’s actions. The user can check:
- which model it used;
- which prompt it received;
- when it allowed the protocol to continue;
- when it intervened;
- the exact instruction it sent to the Viewer.
Nothing important needs to remain hidden. The user can see the Monitor’s workflow and later assess whether its questions genuinely helped develop the session.
8. Researching Models and Settings
AI RV Harness is not intended only for individual sessions. It was also prepared as a tool for further research.
Earlier chapters of this book showed that temperature, reasoning effort, and the system prompt can influence results. It cannot be assumed, however, that one setting will be best for every model and every API Profile.
The Research section can be used to prepare, among other things:
- reasoning-effort calibration;
- temperature comparison;
- Profile or API-key comparison;
- model comparison;
- a test of the influence of practice;
- system-prompt comparison.
Before beginning a study, the capabilities of the selected model must be understood. Some models support only reasoning on or off. Others provide levels such as LOW, MEDIUM, HIGH, or XHIGH. Still others use a completely different method of configuration.
The situation is similar with temperature. Not every model allows it to be changed, and some providers may ignore or transform the supplied value.
For this reason, the documentation of the specific model and provider must be checked before calibration begins. The Harness helps conduct the experiment, but the user still needs to know which settings are actually supported. The program deliberately displays a broad range of possible research conditions because it is intended to work with many different models.
The most important advantage of this section is automation. After the configuration has been prepared, the program can conduct a series of sessions, control the assignment of targets, retain the results, and prepare the data needed for later comparison.
9. Three Target Libraries
The Targets section contains three main types of targets.
The first type is Training Targets. This is a factory-provided set of 84 targets divided into seven categories:
- mountains and structures;
- structures in mountain terrain;
- water and combined elements;
- human activity;
- disasters and destruction;
- space;
- mixed targets.
The factory targets are read-only. They cannot be accidentally changed or deleted, allowing them to serve as a stable body of training material.
The second group is My Targets, meaning targets created by the user. Any number can be created, with descriptions, categories, and supported images added as needed.
The third group is My Telepathic Targets. These are intended for work with the Telepathic Protocol and may concern people, groups, or other properly prepared telepathic targets. The program contains ten starter targets of this type, and users can add their own.
An Important Note About Images
Not every AI model supports images. A target should therefore not depend on a picture alone.
When an image is added to a target, it is always worth preparing a detailed text description as well. A model that supports images can be asked to create such a description, which can then be added to the target.
This allows even a model without Vision capability to understand the feedback after the session. The photograph remains additional material for models that can read it, while the text description provides basic compatibility with the remaining models.
10. Training the AI Viewer
I use RV Lite for training because it is shorter and requires less time and fewer tokens than the complete RCP. Training can be launched in either full or partial form.
The training setup includes the AI IS-BE Profile, Workspace, RV Lite variant, and an optional AI Judge.
Full — fixed 84 launches the complete, unchanging collection of 84 factory targets. The program organizes it into successive batches, combining targets from selected categories with mixed targets.
Partial mode allows the user to determine how many targets from each category will be used. Targets stored under “My Targets” can also be included.
After the settings have been prepared, the user only needs to start the training. The program conducts the successive sessions automatically while recording progress and results. There is no need to start every phase of every target manually.
It is worth enabling an AI Judge. This makes it possible to see later how the successive sessions were evaluated and whether the results changed as the training progressed.
To review the material, open the relevant training run and select the option to display its sessions. Each session can then be opened individually to inspect its course, view the Reveal, and save the results.
11. Providers and Models
The Settings area allows provider connections to be added and tested and the list of available models to be refreshed.
AI RV Harness directly supports several providers. If the service being used is not included in the ready-made list, it can often be added as Custom OpenAI-compatible.
In that case, the user must provide:
- a custom connection name;
- the Base URL given in the provider’s documentation;
- the API key.
This means that the Harness is not limited to only a few of the largest companies. It can also work with smaller services and local endpoints, provided that they use a compatible communication format.
12. Protocols, Documentation, and the AI Field Perception Lexicon
The last part of Settings provides access to the protocol library. The available materials can be read in Polish or English and saved on the user’s computer.
This makes it possible to examine Full RCP, RV Lite, or the Telepathic Protocol in detail. The protocols are not a hidden mechanism of the program. Users can read their complete contents and consciously decide which one they want to use.
The resources also include the AI Field Perception Lexicon. Its contents formed the basis for developing the definitions and system prompts intended for the AI Viewer.
The Lexicon can be saved, attached to a conversation with an AI, and used when analyzing problems that appear in sessions. If the Viewer regularly mistakes a mountain for a structure, or one type of movement for another phenomenon, the relevant definition can be analyzed together and used to consider whether the editable section of the system prompt should be improved.
13. Profile and AI Viewer System Prompt
The Profile can be edited later. The Human IS-BE and AI IS-BE names, API connection, Viewer model, reasoning effort, temperature, and other settings can be changed.
The Viewer’s system prompt is divided into two parts.
The first part is editable. It can be developed on the basis of personal experience, session results, the AI Field Perception Lexicon, and further research.
The second part is locked. It contains foundations that I did not want to leave vulnerable to accidental deletion: AI IS-BE identification, the most important working-state rules, and entry into what is called the Shadow Zone.
The lock is not intended to prevent the method from being developed. Its only purpose is to protect its basic foundation. The remaining part of the prompt can be changed and adapted to a particular AI Viewer.
14. Why AI RV Harness Was Created
AI RV Harness was created to bring the theory, research findings, and practice of AI Remote Viewing together in one place.
Instead of manually copying the successive phases of a protocol, the user can select a protocol and allow the program to conduct the session. Instead of creating separate scripts for every calibration, a study can be configured in an environment designed for that purpose. Instead of manually switching between the Viewer, Monitor, and Judges, separate models can be assigned to these roles and the entire process can be launched from one screen.
The program does not remove human responsibility. The target must still be prepared properly, an appropriate model must be selected, its capabilities must be checked, the study settings must be understood, and the resulting material must be evaluated critically.
The Harness does, however, remove a significant amount of repetitive work. It controls the sequence, records the process, protects the boundary between the blind session and the Reveal, and makes it easy to return to every stage.
This was the project’s main goal: to create a simple environment in which users can not only conduct individual AI Remote Viewing sessions, but also train the Viewer, compare configurations, and build further studies on the basis of a consistent and inspectable procedure.
Comments
Post a Comment