MTP Is a Framework for Tuning LLM Output

It uses sliders, grid coordinates, and presets to shape responses without rewriting the task itself.

MTP (Mapping the Prompt) is a framework for steering LLM output with grids and sliders instead of long natural-language behavior instructions. It is designed to make the ideas and concepts in a prompt easier to express intuitively, helping the user and the LLM align with fewer instructions.

Its core is a 3x3 color arrangement made of nine nodes. The relationships between color, position, polarity, and intensity are defined from this arrangement. Each color in the 3x3 arrangement defines nodes for the Side A and Side B polarities. For example, Yellow in the upper-left corresponds to Open on Side A and Still on Side B.

Diagram showing the MTP coordinate system and node layout. The left side shows the relationship between Side A / Side B colors and nodes, and the right side shows examples of coordinate pairs such as D:4 and A:19.

The column:row color grid on the right shows an image for unifying Side A and Side B nodes and treating them as coordinates in a UI. It shows an example using the D:4 and A:19 coordinates.

MTP Skill is an Agent Skill for using the MTP framework through the /mtp command.

What is MTP?

Traditional prompts often adjust model behavior through natural-language instructions like these:

Act as an expert.
Be more concise.
Think step by step.

MTP moves that behavioral steering into non-verbal parameters such as color, position, and intensity. It does not change the task itself; it adjusts qualities such as force, flow, depth, structure, openness, and focus.

Three ways to control output

MTP Skill supports three input modes. Internally, each mode resolves into the same form: axis, polarity, and intensity.

Mode	Pattern	Use it when
Slider	`power:100`, `flow:70`, `haze:50`	You need explicit controls that are easy to read.
Grid	`J:4`, `D:16`, `A:1`	You want compact steering through coordinates.
Preset	`strategist`, `synthesizer`, `maverick`, `concierge`	You want a reusable blend of multiple coordinates.

# Slider
/mtp power:100 Summarize this article

# Grid
/mtp J:4 Explain this concept

# Preset
/mtp strategist Compare these options

Slider and Grid UI Preview

The MTP Interactive UI on the roadmap is planned to make sliders and grids available as visual controls. The UI is not implemented yet; at the current stage, MTP Skill is operated through the /mtp command.

The images below show a UI preview of sliders for controlling intensity and a grid for controlling position. They make it easier to see how coordinates such as J:4 and D:16 correspond to color and position. The slider view and the grid view both represent the same node system from different angles.

Slider

Diagram explaining MTP slider arguments using /mtp node. The left panel shows node sliders for Side A and Side B with Power set to 70. The right panel shows node intensity as 3D Chebyshev distance, with labeled directions such as Open, Power, Focus, Flow, Close, Surge, Collapse, and Fade. Slider arguments combine a node name with an intensity value, moving from the center toward a Side A or Side B direction in the MTP space.

Grid

Diagram explaining MTP grid arguments using /mtp column. The left panel shows a simplified 3×3 color grid projected onto a 19×19 coordinate grid, with example coordinates such as J:4, D:10, P:10, and J:16. The right panel shows RGBA color values across the full 19×19 grid. Grid arguments select a point on the 19×19 MTP coordinate plane, where positions map back to node directions and their RGBA color values.

In the grid UI, position is intended to be selected as if placing a point on the grid.

Go to the MTP Skill documentation →