Skip to main content

Fragmented Consciences

Fragmented Consciences & the Architecture of Distributed AI Reasoning​

(This file includes an embedded Mermaid diagram at the end.)

1. Overview​

Large Language Models are powerful reasoning engines, but they are:

  • stateless
  • constrained by context window limits
  • blind to long-term information unless explicitly provided
  • unable to think across time or large datasets without orchestration
  • limited to single-turn local reasoning

Brainframe overcomes these limits through Fragmented Consciences, Context Buckets, Reasoning Pipelines, and Orchestration.

2. Fragmented Consciences​

A “Fragmented Conscience” is a specialised reasoning module dedicated to one domain:

  • Risk
  • Programs
  • Incidents
  • Transport
  • Scheduling
  • Policy
  • Staff welfare
  • etc.

Each conscience:

  1. Ingests data
  2. Processes & reasons
  3. Stores its current “state of insight”

3. The Three Layers of Reasoning​

3.1 Raw Data​

3.2 Processed Data (maths, filters, clustering)​

4. Context Buckets​

Each conscience stores its results in a bucket:

  • Raw input
  • Processed results
  • AI reflections
  • Multi-size summaries
  • Deltas
  • Relevance weights

5. Why Fragmentation Matters​

The LLM context window is tiny.
The mind must not live inside the LLM.
Brainframe sends only the relevant slice of the mind each turn.

6. The Goo Ball​

The Goo Ball contains:

  • Conversation Goo
  • User Goo
  • Agent Goo
  • Data Goo

7. Big Question Answering​

When asked a complex question, Brainframe:

  1. Classifies the question
  2. Activates relevant consciences
  3. Assembles a Goo Ball from their insights
  4. Calls the LLM
  5. Produces a final, deeply informed answer

8. Why This Works​

LLM ≠ brain
Brainframe = brain
Together = cognition

9. Diagram​

flowchart TD

UserQuery([User Question])

subgraph Brainframe
    QC[Question Classifier]
    G1[Conscience: Risk]
    G2[Conscience: Programs]
    G3[Conscience: Incidents]
    G4[Conscience: Transport]
    G5[Conscience: Scheduling]
    AG[Assemble Goo Ball]
end

LLM[LLM Model]

UserQuery --> QC
QC --> G1
QC --> G2
QC --> G3
QC --> G4
QC --> G5
G1 --> AG
G2 --> AG
G3 --> AG
G4 --> AG
G5 --> AG
AG --> LLM
LLM --> Out[Final Answer]

10. Final Definition​

Fragmented Consciences are specialised, continuously running AI subsystems that maintain domain knowledge so the LLM sees a prepared mind, not raw chaos.