Axis Thought Packet

A coordinate system for AI self-awareness — knowing where you are in thought space.

Axis Thought Packet is a structured thought-state representation format for human-AI and multi-agent collaboration. Model-agnostic: works with Claude, GPT, Gemini, or any LLM.

---

Install

pip install axis-thought-packet

---

The Problem

Every AI session starts from zero.
You re-explain context. The AI re-builds understanding. Thought state is lost.

Axis Thought Packet solves this by treating thought as a portable, structured object — with coordinates, history, and intent — that can be passed between sessions, agents, and models.

---

Quick Start

from axis_thought_packet import ThoughtPacket

# Create a thought packet
packet = ThoughtPacket(
    focus="Build a model-agnostic thought protocol",
    intent="Enable AI self-positioning across sessions",
    confidence=0.85,
    subject="Axis Thought Packet",
    context="Multi-agent AI collaboration"
)

# Where am I in thought-space?
print(packet.position)
# → {'clarity': 1.0, 'confidence': 0.85, 'novelty': 0.5, 'completeness': 0.75}

# Generate a resumption prompt — paste into ANY AI
print(packet.to_prompt())
# → ## Axis Thought Packet — Context Handoff
#   **Focus**: Build a model-agnostic thought protocol
#   **Intent**: Enable AI self-positioning across sessions
#   ...

# Save and reload across sessions
packet.save("my_session.atp")
reloaded = ThoughtPacket.load("my_session.atp")

---

Core Concepts

Universal Formula

Axis Thought Packet is based on the Universal Formula framework, which treats thinking as three operations:

Thought = State × Transition × Translation

Component	Description	In code
State	What the AI knows and intends right now	thought_state
Transition	How the state is evolving	transition
Translation	How to hand off to another agent or session	to_prompt()

Thought Position

Every packet has a position in thought-space — a coordinate vector that answers: "Where am I right now as a thinking agent?"

packet.position
# {
#   "clarity":      0.85,  # how well-defined is the focus?
#   "confidence":   0.90,  # how confident is the intent?
#   "novelty":      0.50,  # how unexplored is the territory?
#   "completeness": 0.75   # how much context is filled in?
# }

This is the key insight: AI can know where it is in thought-space, not just what it's saying.

---

Features

ThoughtPacket — Core object

from axis_thought_packet import ThoughtPacket

packet = ThoughtPacket(
    focus="...",          # what are we thinking about?
    intent="...",         # why are we thinking about it?
    confidence=0.8,       # how sure are we?
    subject="...",        # topic domain
    context="...",        # situational context
    stance="...",         # current stance or angle
    to_state="...",       # where are we headed?
    tags=["research"]     # free-form labels
)

ThoughtLens — Perspective transformer

View the same thought through different lenses:

from axis_thought_packet import ThoughtLens

critic   = ThoughtLens("critic")       # reduces confidence, adds scrutiny
engineer = ThoughtLens("implementer")  # shifts toward practical execution
user     = ThoughtLens("user")         # reframes around usability

critical_view = critic.apply(packet)
print(critical_view.thought_state.confidence)  # lower than original

Built-in lenses: critic, implementer, user, summarizer

Multi-agent handoff

from axis_thought_packet.utils import merge_packets

agent_1 = ThoughtPacket(focus="Defined schema", intent="Establish format", confidence=0.9)
agent_2 = ThoughtPacket(focus="Wrote tests",   intent="Verify behavior",  confidence=0.85)
agent_3 = ThoughtPacket(focus="Shipped v0.1",  intent="Public release",   confidence=0.95)

final = merge_packets(agent_1, agent_2, agent_3)
# → merged focus, averaged confidence, combined tags

Thought evolution diff

from axis_thought_packet.utils import diff_packets

before = ThoughtPacket(focus="Idea stage",          confidence=0.4)
after  = ThoughtPacket(focus="Implementation stage", confidence=0.85)

changes = diff_packets(before, after)
# → {'thought_state.focus': {'before': 'Idea stage', 'after': 'Implementation stage'}, ...}

---

Why This Matters

Most AI frameworks focus on pipelines (what the AI does).
Axis Thought Packet focuses on state (where the AI is).

Framework	Focus
LangChain / LlamaIndex	Task pipelines
AutoGen / CrewAI	Agent orchestration
Axis Thought Packet	Thought state representation

These are complementary, not competing. ATP is the cognitive layer that sits beneath any pipeline.

Inspired by:

• EML functions — any elementary function from a single binary operator
• Fourier analysis — any waveform from simple basis functions
• HFE (Hypothesis Field Engine) — reasoning as geometric navigation in meaning space
• The Universal Formula — thought as state, transition, translation

---

Repository Structure

axis-thought-packet/
├── src/axis_thought_packet/   # Python package
│   ├── packet.py              # ThoughtPacket core
│   ├── lens.py                # ThoughtLens transformer
│   └── utils.py               # merge, diff, load, save
├── tests/                     # 13 tests, all passing
├── examples/                  # quickstart.py
├── schema/                    # JSON Schema spec
├── docs/                      # specification notes
└── paper/                     # manuscript drafts

---

Citation

If you use Axis Thought Packet in research, please cite:

@software{axis_thought_packet_2026,
  author    = {Da-P-AIP},
  title     = {Axis Thought Packet: A structured thought-state representation
               format for human-AI and multi-agent collaboration},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20116884},
  url       = {https://github.com/Da-P-AIP/axis-thought-packet}
}

Citation metadata: CITATION.cff · Zenodo metadata: .zenodo.json

---

License

This repository uses a dual-license structure:

• Source code in src/, tests, and executable examples are licensed under the MIT License.
• Documentation, schemas, research notes, and publication materials are licensed under Creative Commons Attribution 4.0 International (CC-BY-4.0) unless otherwise noted.

See LICENSE.md.