AXON: File-Based AI Encyclopedia

AXON is a transparent, file-based AI system that implements a knowledge base through the 9P protocol. It embodies the Plan 9 philosophy that "everything is a file" — including AI reasoning, knowledge storage, and multi-agent consensus building.

Core Concepts

AXON treats AI operations as filesystem operations. You search, query, and manage knowledge by reading and writing files. The system provides:

• 9P Filesystem interface: Complete AI system exposed as a mountable filesystem
• Multi-mind fact extraction: Multiple AI agents (literal, skeptic, synthesizer, pattern_matcher, questioner) analyze facts independently
• Consensus building: Contradiction detection and confidence scoring across multiple minds
• Plan 9-style inbox processing: Drop files in /inbox and trigger processing via /ctl
• Memory systems: Episodic, semantic, and procedural memory storage
• Transparent reasoning: All AI operations visible through file operations

Filesystem Structure

# AXON filesystem layout
/mnt/axon/
├── ctl                    # Control file - trigger actions
├── encyclopedia/
│   ├── entries/          # Knowledge base entries
│   └── search/
│       ├── query         # Write search terms here
│       └── results       # Read results here
├── inbox/
│   └── incoming/         # Drop files for processing
├── minds/                # Multi-mind status
│   ├── status
│   ├── literal/
│   ├── skeptic/
│   └── synthesizer/
└── memory/
    ├── episodic/
    ├── semantic/
    └── procedural/

Quick Start

# Build and start AXON
$ cd axon
$ mk axon-build
$ ./bin/axon -p 17020

# Mount the 9P filesystem
$ 9p -a localhost!17020 mount /mnt/axon

# Search the knowledge base
$ echo "sound waves" > /mnt/axon/encyclopedia/search/query
$ cat /mnt/axon/encyclopedia/search/results

# Process files through inbox
$ cp email.txt /mnt/axon/inbox/incoming/
$ echo 'process_inbox' > /mnt/axon/ctl

# Check mind status
$ cat /mnt/axon/minds/status
$ cat /mnt/axon/minds/literal/status

Multi-Mind Architecture

AXON uses multiple specialized "minds" (AI agents) to extract and verify facts:

• Literal: Direct fact extraction from source text
• Skeptic: Questions assumptions and identifies potential contradictions
• Synthesizer: Combines multiple facts into coherent understanding
• Pattern Matcher: Identifies patterns and relationships across entries
• Questioner: Generates relevant questions to guide further exploration

Each mind operates independently and reports to shared consensus files, enabling transparent reasoning and contradiction detection.

Links

GitHub Repository | Back to Projects