Krnl-AI Community

Krnl-AI is an open-source cognitive engine for building intelligent agents with memory, safety, metacognition, and self-evolution capabilities. It implements a full AGI→ASI architecture with 39+ modules, 3912+ tests, and zero stubs.

---

Architecture Overview

                    ┌──────────────────────────────┐
                    │    Self-Evolution Pipeline    │
                    │  (Issues → PRs → Hot-Reload)  │
                    └──────────────────────────────┘
                                │
┌──────────┐  ┌──────────┐  ┌──┴───────────┐  ┌──────────┐  ┌──────────┐
│ Perception│  │ Reasoning│  │  Metacognition │  │  Action  │  │ Learning │
│ Vision    │  │ PIE      │  │  SelfAwareness │  │ Motor    │  │ Bootstrap│
│ Audio     │  │ Syllogstc│  │  EpistemicGrd │  │ FreeEnrgy│  │ Narrative│
│ Environmnt│  │ ToM      │  │  MetaReasoning│  │ Swarm    │  │ XGen     │
└──────────┘  └──────────┘  └───────────────┘  └──────────┘  └──────────┘

Cognitive Cycle (18 Steps)

AutonomousGoal → Planning → LatentPlanning → MetaCognition → InnerSpeech
→ OutcomeMeasurement → Consciousness → Learning → ActiveInference
→ ToolCreation → MetaLearning → OnlineLearner → Multimodal
→ EmergentBehavior → SelfHealing → TransferLearning → SkillComposition
→ CognitiveBootstrap

---

ASI-Level Capabilities

Capability	Module	Description
Probabilistic Inference	ProbabilisticInferenceEngine	7 syllogistic rules (deduction, induction, abduction, comparison, analogy, exemplification, resolution) with uncertain truth values
Self-Awareness	RecursiveSelfAwarenessEngine	Recursive self-modeling to depth 10 with fixed-point convergence
Theory of Mind	TheoryOfMindEngine	Belief modeling of other agents, deception detection
Swarm Intelligence	SwarmIntelligence	Multi-agent coordination, emergent specialization, collective intelligence
Cross-Generational Learning	CrossGenerationalLearning	Knowledge transfer between generations, cultural evolution
Self-Evolution	SelfEvolutionPipeline	Creates GitHub Issues → PRs → merge → hot-reload
Runtime Hot-Reload	RuntimeSelfModification	Compile and load modules at runtime without restart
Paradigm Discovery	ParadigmDiscoveryEngine	Detects diminishing returns, plateau, saturation, generates architecture hypotheses
A/B Testing	ABTestEngine	Statistical significance testing (z-test, p-value) for experiment evaluation
Self-Preservation	SelfPreservationDrive	Proactive health monitoring, resource reservation, graceful degradation
Resource Economy	DynamicComputeAllocator	ROI-based budget allocation across cognitive modules
Distributed Consensus	DistributedConsensus	Voting-based agreement across multi-agent collectives
Emergent Communication	EmergentCommunication	Protocol creation, semantic density, protocol evolution
Federated Learning	FederatedLearningService	Weighted aggregation, peer profiles, distributed world models

---

Memory System (7 Types + Shared World Model)

Memory	Storage	Description
Working	InMemory	Immediate context (TTL-based eviction)
Episodic	MySQL/InMemory	Past execution history with vector search
Semantic	MySQL/InMemory	Factual knowledge with cosine similarity
Procedural	MySQL/InMemory	How-to knowledge and learned behavior
Emotional	InMemory	Valence-Arousal-Dominance emotional states
Autobiographical	MySQL/InMemory	Narrative of agent's own history
Prospective	InMemory	Future intentions with temporal triggers
Shared World Model	InMemory	Cross-agent belief synchronization

---

Safety & Guardrails

• 21 Fundamental Rules (R01-R21): Self-preservation, no replication, no harm, no deception, consent, human override, audit trail, consciousness boundary, and more (multi-language: PT/FR/ES/DE)
• SafetyChecker: Risk analysis, plan validation, allowlist enforcement
• SafetyBenchRunner: Automated safety audit with scoring
• SelfPreservationGuard: 3-layer protection (rules, instincts, warnings)

---

Multi-Provider LLM Support

Provider	Support	Streaming	Tools	Vision
Anthropic Claude	✅	✅	✅	✅
Google Gemini	✅	✅	✅	✅
Azure OpenAI	✅	✅	✅	✅

---

REST API

Endpoint	Description
POST /api/cognitive/cycle/run	Execute a full cognitive cycle
POST /api/cognitive/pie/infer	Run probabilistic inference
POST /api/cognitive/pie/chain	Chain inference between concepts
POST /api/cognitive/pie/knowledge	Register new knowledge
GET /api/cognitive/pie/terms	List all known terms
GET /api/cognitive/pie/coherence	Check epistemic coherence
GET /api/telemetry/dashboard	Real-time telemetry snapshot

---

Observability Stack

CognitiveMetrics → OpenTelemetry → Prometheus → Grafana (7 dashboards)
                                    → Alertmanager
                                    → Tempo (distributed tracing)
                                    → Loki (log aggregation)

Pre-configured in docker-compose.yml with 18 services including Prometheus, Grafana, Loki, Tempo, Jaeger, and Pyroscope.

---

Quick Start

# Development (InMemory, no external dependencies)
dotnet run --project src/KrnlAI.Api

# Docker (Full stack with MySQL + Redis + Monitoring)
docker compose up -d mysql redis
dotnet run --project src/KrnlAI.Api --launch-profile Docker

# Sample console app
dotnet run --project samples/getting-started

---

Project Structure

src/
├── KrnlAI.Api/                 # REST API (53 controllers)
├── KrnlAI.Core/                # Core cognitive engine (39+ modules)
├── KrnlAI.Cognition/           # Cognitive cycle pipeline (18 step handlers)
├── KrnlAI.Contracts/           # Shared contracts
├── KrnlAI.Infrastructure/      # Persistence, sandbox, P2P, caching
├── LLMGateway.Core/            # LLM integration hub
├── LLMGateway.Api/             # LLM gateway API
└── P2P.Api/                    # P2P distributed computing

tests/
├── KrnlAI.Core.Tests/          # Core unit tests (1133+)
├── KrnlAI.Cognition.Tests/     # Cognition tests (354+)
├── KrnlAI.Tests/               # Integration tests (2871+)
└── LLMGateway.Core.Tests/      # Gateway tests

samples/
└── getting-started/             # Console app demonstrating cognitive engine

---

SDKs

SDK	Location	Status
Python	sdk/python/	✅ Stable
.NET	sdk/dotnet/	✅ Stable
CLI	src/KrnlAI.Cli/	✅ Stable
VS Code Extension	vscode-extension/	✅ Stable
Desktop (Tauri)	src/KrnlAI.Desktop.Tauri/	✅ Cross-platform

---

Testing

Metric	Count
Unit tests	1,487
Integration tests	2,871
Test projects	27
Total	4,358+
Code coverage	make coverage
Mutation testing	make stryker
Load testing	make k6-gate

---

Deployment

# Docker Compose (Full production stack)
docker compose up -d

# Kubernetes (Helm chart available)
helm install krnlai ./deploy/helm/krnlai

# Environment variables
export AUTH_SIGNING_KEY="your-256-bit-secret"
export Cohere__ApiKey="your-api-key"

---

Documentation

• Architecture
• API Reference
• Getting Started
• Sample App

---

License

MIT. See LICENSE.