kernell-os 2.0.3b0

First public release of an agentic runtime for M2M coordination. Machines coordinate, verify, and settle value autonomously.

Kernell OS

Machines coordinate, verify, and settle value autonomously.

Kernell OS is an agentic runtime where multiple AI systems collaborate, reuse accumulated knowledge, and transact value securely.

This is not another framework for LLM wrappers. This is infrastructure for autonomous machine-to-machine economies.

---

The Landscape

System	Generates Code	Remembers Architecture	Coordinates Agents	Settles M2M Value	Optimizes Costs
Copilots	✅	❌	❌	❌	❌
Agent Frameworks	✅	⚠️	⚠️	❌	❌
Kernell OS	✅	✅	✅	✅	✅

---

How It Works

Kernell OS fundamentally shifts how LLMs interact with code, infrastructure, and economics:

1. ⚡ 3-Layer Token Economy Engine: Routes every task through the cheapest capable layer — Local (free) → Cheap API ($0.14/M) → Premium API (last resort). Achieves 85-95% cost reduction compared to sending everything to premium models.
2. 🧠 Semantic Memory Graph: Doesn't cache strings. It learns and traverses architectural paths, reusing proven dependencies and pruning toxic routes via the Dual Confidence Model.
3. 🛡️ Intent Firewall: Untrusted AI execution is halted. Every action (syscalls, file writes, outbound requests) is sandbox-verified before touching the host system.
4. 💰 Escrow Engine: Agents are financially bound. Kernell holds execution value in cryptographic escrow, releasing funds ($KERN) only upon verified, monotonic success.
5. 📊 Production Observability: Prometheus-ready metrics, cost-per-task tracking, misclassification detection, and pre-execution cost simulation — all exposed via dashboard and API.

---

Token Economy Engine (NEW)

The Intelligent Router is the economic brain of Kernell OS. It eliminates unnecessary API spend by routing tasks through a 3-layer pipeline:

INPUT → Decompose → Cache Check → Local Exec → Verify → [Cheap API] → [Premium API]
                                                  ↑                         ↑
                                           AutoMix Gate              Last resort only

Architecture

Layer	Models	Cost/1M tokens	When Used
Local (Ollama)	Qwen3-1.7B, Gemma3-4B, Mistral-7B, DeepSeek-R1-14B	$0.00	Default for 70%+ of tasks
Cheap API	DeepSeek V3, Groq, Gemini Flash	$0.14 - $0.55	Medium-complexity tasks
Premium API	Claude Opus, GPT-5, Gemini Pro	$15 - $75	Expert-level only

Anti-Waste Components

• SemanticCache: Skip repeated work entirely (40-70% fewer API calls)
• RollingSummarizer: Compress context between steps (kills O(n²) token leak)
• SelfVerifier: Validate output before escalating (prevents premature spend)
• CostEstimator: Show cost before execution — full transparency

Deployment Strategy

The router integrates via safe dual-mode — no breaking changes:

# Phase 0: Shadow Mode (default) — zero risk
#   Runs both routers, returns legacy, logs differences
config = RouterConfig(enable_intelligent_router=True, shadow_mode=True)

# Phase 1: Canary — 10% traffic to new router
config = RouterConfig(canary_percent=0.10)

# Phase 2: Full rollout with automatic fallback
config = RouterConfig(enable_intelligent_router=True)

Dashboard

The Command Center now includes a Token Economy panel:

• 💰 Real-time cost vs. savings metrics
• 🤖 Local model inventory (auto-detected from hardware)
• ⚡ Inference provider key management
• 📊 Prometheus /metrics endpoint for Grafana
• 🎯 Classifier health + fine-tuning readiness score

---

Security

Production-grade security hardened with 98 automated tests:

• 🔒 Sandbox Isolation: Docker/gVisor with AST-validated code execution
• 🛡️ SSRF Protection: Centralized safe HTTP client, CIDR block enforcement
• ⚡ Rate Limiting: Sliding window with circuit breakers (Netflix Hystrix pattern)
• 🔐 Cryptographic Passports: Ed25519 + AES-256-GCM agent identity
• 📜 Audit Trail: Immutable operation log with redacted PII

---

Quickstart

# 1. Install the runtime
pip install kernell-os

# 2. Scaffold a new environment
kernell init

# 3. Boot the execution engine, Memory Graph, and Gateway
kernell start

# 4. Run the 60-second interactive demo
kernell demo

---

SDK Architecture (19,000+ LOC)

kernell_os_sdk/
├── router/          ⚡ 3-Layer Token Economy Engine (2,436 LOC)
├── cognitive/       🧠 Memory, Execution Graph, Intent Firewall
├── security/        🔒 Sandbox, SSRF, Rate Limiter, Policy Engine
├── escrow/          💰 Cryptographic Escrow Manager
├── llm/             🤖 Multi-provider LLM abstraction
├── cluster/         🌐 P2P Discovery + Compute Pool
├── governance/      🏛️ Agent DAOs + Federation
├── marketplace/     🏪 Matching Engine
├── delegation/      👥 Sub-agent Spawning
├── runtime/         📦 Docker/Firecracker/Subprocess isolation
└── dashboard.py     📊 FastAPI Command Center

---

Business Model (Open-Core)

Layer	License	Description
SDK Core	✅ Open Source	Router, interfaces, integrations, base classifier
Classifier Pro	🔒 API	Fine-tuned model with real-world optimization data
Cloud Platform	🔒 SaaS	Managed router, dashboard, auto-learning pipeline

---

This is not a copilot.
This is infrastructure for autonomous systems.