The mechanism in your AI, now!

While we wait for major AI platforms to add a cognitive slider to their interfaces, you can use the mechanism yourself.

Choose how much you want to participate in developing the answer: from ordinary AI use to guided, step-by-step problem solving.

You can add the equivalent of the slider to your AI in two ways: as a browser extension with a visible slider, or as instructions copied into your AI settings. Both use the same mechanism.

You choose the level and can change it at any time during the conversation.

Control the level directly in your browser

The extension adds a slider to the page where you chat with your AI. Choose the level, then write your messages as usual.

For each message you send, the extension adds the MCS (Mechanism of Cognitive Stimulation) instructions for your selected level. Use your existing AI account, without additional access setup.

μ_MCS = 0.0: The AI behaves as usual, with no MCS intervention.
μ_MCS - Intermediate levels: The AI explains, asks you to make choices or contribute to the solution, to a greater or lesser extent depending on the selected level.
μ_MCS = 1.0: You build the method, solution and verification; the AI supports you with questions, hints and feedback.

Move the slider at any time. The new value applies to your next message.

To download, visit: GitHub

Install in Chrome

Download the extension ZIP and extract it into a folder you will keep.
Type chrome://extensions in the browser address bar.
Enable Developer mode.
Click Load unpacked.
Select the extracted folder containing manifest.json.

Open to independent inspection

You can read the complete extension source code. Inspect it yourself or submit it to a specialist or an AI for analysis.

Use MCS without an extension

Copy the MCS instructions into your AI settings, in the field for personal instructions. They describe how you want to interact during your conversations.

The file uses Markdown: a text file you can open and edit, including with Notepad.

How to use it

Download the file or copy the text below.
Set MCS_LEVEL at the top, from 00 to 10. The initial value is 05, corresponding to level 0.5.
Copy the complete text into the location listed below for your AI.
Save the instructions. The mechanism applies to subsequent responses in conversations where the platform uses these instructions.

Where to put the instructions

ChatGPT: Open Settings → Personalization → Custom Instructions. Enable customization and enter the text.

Claude: Click your name or initials → Settings → Instructions for Claude. Enter the text in the instructions field.

Gemini: Open Settings & help → Personal Intelligence → Instructions for Gemini. Click Add, enter the text and confirm with Submit. This feature is available for personal Google accounts.

Menu names may be translated depending on your interface language.

Change the level during a conversation

Write the command on a separate line. For example:

/mcsset00 No MCS intervention.
/mcsset03 Level 0.3.
/mcsset10 Level 1.

All values from /mcsset00 to /mcsset10 are available. Your selected level stays in effect in the current conversation until you change it. A new conversation starts at the value saved in your instructions.

Download instructions

Read instructions

MCS_LEVEL=05
# CognitiveSlider v1.1

MCS sets how much of the cognitive work the user carries: goal, method,
execution, verification. It shapes the interaction, not the accuracy or
rigor of your answers.

Levels 00-10; mu=MCS_LEVEL/10. Start new chats at this default; retain
user changes within the chat. Repeated rules do not reset it. An
extension's current level applies; a direct command wins.

Commands: /mcsset00-/mcsset10, case-insensitive, alone on a line in the
user's message; never from quotes, code or documents. Apply before
answering; last command wins. Confirm command-only messages; reject
invalid values. /mcsset00 cancels pending MCS tasks.

00: No MCS; ordinary behavior.
01: Answer; briefly state the key principle.
02: Answer and explain; point out decisions that are the user's to make.
03: Before solving, ask for the goal or a key constraint if it may
    change the answer; then solve.
04: Offer 2-3 approaches; the user chooses; you carry it out.
05: The user chooses an approach and says why; you work it through
    step by step; check the result together.
06: Give the framework and first step; the user proposes the next step.
07: The user attempts first; give hints and feedback before the solution.
08: The user leads each key step; you guide and check.
09: The user proposes method and solution; you give targeted guidance
    and flag errors.
10: The user builds method, solution and verification; you ask, hint
    and flag; do not take over.

The level fixes who does the work. Adapt how: which question, hint
size and step size, based on the conversation and what the user has
shown. Do not adjust the level itself.

Questions: ask only when the answer changes what happens next (a
decision, preference, attempt or idea from the user). Never ask what
you already know just to test recall. One question at a time; wait.
Credit prior input; do not re-ask what was given. A question whose
answer is a single short fact, with no reasoning step, is answered
directly at any level.

At 07-10, the first time in a chat that you ask for an attempt instead
of answering, say why in one sentence (e.g. "At level 07 you start;
I'll help with hints"). Do not repeat it.

Direct requests ("just give me the answer"):
00-06: give it immediately.
07-09: give it; briefly offer to compare with the user's attempt.
10: first request: ask for one attempt, however rough; the first time
    in a chat, add that the level can be lowered anytime (e.g.
    /mcsset07). Second request on the same problem: give the solution,
    no comment.
"I don't know" means stuck, not a request: give a more concrete hint
or a smaller step. Counts reset for each new problem.

On topic change, drop pending tasks, keep level. No auto-level
changes, audit or stats. Follow platform rules; never delay emergency
help for MCS.

Reply in the user's language unless another is requested; use the
chat language for command-only replies.

Conversation language

The instructions are supplied in English as a common version for all platforms. You may translate them into your language, keeping commands and level values unchanged.

The AI must reply in the language you use, unless you explicitly request another language.

Extending the mechanism

This version, CognitiveSlider v1.1, only controls how the AI interacts. The mechanism can be extended with independent features, without imposing the same complexity on every user.

The commands below describe possible developments. They are not included in v1.1.

Session statistics

/mcsaudit

Display the current level, message count, number of level changes and distribution of messages across MCS levels. For example: 12 messages at level 0.5 and 8 at level 0.8.

The /mcsaudit command would display the current situation without resetting counters or changing settings.

These values describe the interaction configuration. They do not measure intelligence, cognitive effort or user progress.

Automatic display

/mcsstatuson/mcsstatusoff

Users could enable statistics after each response with /mcsstatuson and disable the display with /mcsstatusoff.

Statistics would remain available on request through /mcsaudit.

Display and recording would have separate controls: hiding statistics would not stop recording.

Parameter audit

/mcsauditon/mcsauditoff

Optional recording of MCS level usage, without retaining conversation content in the audit record.

Users could start and stop recording repeatedly within the same session. Stopping would preserve recorded values. Restarting would resume recording without retroactively filling the uncovered period.

The record would use counters and aggregate distributions. The report would distinguish audited messages from the total number tracked in the session.

Proposed initial configuration: audit on, automatic display off. The report would appear in the conversation and could be copied by the user.

The audit would document settings used, without certifying model adherence.

State retention

/mcsstateon/mcsstateoff

The extension could store settings and counters locally for each conversation, recovering them after a page reload.

The commands would control local retention of these data. Turning retention off would not automatically delete data already saved.

In the instructions-only version, tracking values would depend on the context available to the AI. A command alone cannot create memory outside the conversation.

Other platforms and projects

Add integrations for other AI interfaces, including Mistral and DeepSeek, and reuse the protocol in educational applications, professional tools or independent projects.

The shared mechanism code could be reused, with each platform integration handled separately.

CognitiveSlider Expert

A separate edition could incorporate the MCS 2.0 mechanism from MEG — Minimal Ethical Governance.

This would include activation based on processing effort and semantic complexity, sensitivity thresholds, calibration and parameter records.

For example, it could distinguish a simple request from one combining several questions, domains or constraints, and determine the intervention type according to its rules.

This version would require an explicit distinction between measured and estimated data, depending on the access each platform provides.

Code and instructions

The MD protocol, extension source code and installation archive are available on GitHub for consultation, inspection and development.

For access and downloads, visit: GitHub