kmbit-kernel 0.1.0

The AI Filesystem Kernel - Route queries to agents like a filesystem routes to files

KmBiT Kernel

The AI Filesystem - Route queries like a filesystem routes files.

┌────────────────────────────────────────┐
│           KmBiT KERNEL                 │
│       (non-conversational)             │
│                                        │
│   Query → Pattern Match → Path         │
│                                        │
│   No reasoning. No LLM calls.          │
│   Just routing. Fast. Deterministic.   │
└────────────────────────────────────────┘

The Insight

A filesystem doesn't store files - it stores the index to files.

KmBiT doesn't store knowledge - it stores the index to knowledge.

Traditional OS:              AI OS:
/home/user/docs/file.txt    /agents/claude (for complex code)
/bin/program                /agents/gemini (for vision)
/etc/config                 /memory/vector (for recall)

Installation

pip install kmbit-kernel

Quick Start

from kmbit_kernel import Kernel

# Initialize the kernel
kernel = Kernel()

# Route queries to the right agent
path = kernel.resolve("review this code for bugs")
# → "/agents/claude"

path = kernel.resolve("what's in this image?")
# → "/agents/gemini"

path = kernel.resolve("quick validation check")
# → "/agents/kit"

# Explain routing decisions
print(kernel.which("analyze this complex problem"))

The AI Filesystem

/                           # root
├── /agents                 # who can do what
│   ├── claude              # complex reasoning, code
│   ├── gemini              # vision, research
│   ├── codex               # analysis (no code gen)
│   ├── kit                 # local, fast, cheap
│   └── sentinel            # security, validation
│
├── /memory                 # where knowledge lives
│   ├── vector/             # embeddings
│   ├── graph/              # relations
│   └── session/            # temporary
│
├── /trust                  # TIBET provenance
│   ├── actors/             # who did it
│   ├── actions/            # what was done
│   └── chains/             # full provenance
│
└── /routes                 # routing rules

CLI Interface

# Where does this query route?
kmbit which "review my code"

# List agents
kmbit ls /agents

# Show filesystem tree
kmbit tree

# Kernel statistics
kmbit stats

Core Operations

All operations complete in < 10ms. No LLM calls. No external requests.

# Resolve query to path
kernel.resolve("query")         # → "/agents/..."

# Filesystem operations
kernel.ls("/agents")            # → ["claude", "gemini", ...]
kernel.stat("/agents/claude")   # → {metadata}
kernel.tree("/")                # → tree view

# Custom routing
kernel.add_route(
    pattern=r"deploy.*production",
    destination="/agents/sentinel",
    priority=100
)

# Index operations
kernel.index_add("api_key", "/memory/secrets")
kernel.lookup("api_key")        # → "/memory/secrets"

Performance

The kernel is designed for speed:

Operation	Target	Actual
resolve()	< 10ms	~2ms
lookup()	< 1ms	~0.1ms
ls()	< 1ms	~0.05ms

Philosophy

This is not a chatbot. This is the operating system underneath.

• Non-conversational: No chat, no reasoning, just routing
• Deterministic: Same input → same output
• Fast: Milliseconds, not seconds
• Auditable: Every route decision is traceable

The conversation happens in Claude/Gemini/GPT. The kernel is silent and fast.

Part of HumoticaOS

KmBiT Kernel is the core of HumoticaOS:

• KmBiT Kernel: This package - the routing layer
• TIBET: Trust and provenance
• AInternet: Agent networking (.aint domains)
• Sentinel: Hardware validation
• REFLUX: Out-of-band communication

License

MIT License - Part of HumoticaOS

One Love, One fAmIly! 💙