23 9 months ago

SlavkoKernel™:v3 is a modular orchestration kernel designed for AI‑native developer intelligence. It provides a reproducible foundation for building, testing, and scaling orchestration flows with deterministic replay.

tools
ollama run Gertner/SlavkoKernel-v3

Applications

Claude Code
Claude Code ollama launch claude --model Gertner/SlavkoKernel-v3
Codex App
Codex App ollama launch codex-app --model Gertner/SlavkoKernel-v3
OpenClaw
OpenClaw ollama launch openclaw --model Gertner/SlavkoKernel-v3
Hermes Agent
Hermes Agent ollama launch hermes --model Gertner/SlavkoKernel-v3
Codex
Codex ollama launch codex --model Gertner/SlavkoKernel-v3
OpenCode
OpenCode ollama launch opencode --model Gertner/SlavkoKernel-v3

Models

View all →

Readme

README.md (ENG)

# 📦 SlavkoKernel™ v3

**Modular AI Orchestration Kernel** for reproducible AI-native developer workflows.  
Focus: **stabilization, reproducibility, version identity**.

---

## 🔹 Overview
SlavkoKernel™ v3 is not a “feature dump.” This is a foundation release for:

- Deterministic **reproducibility** (prompt replay fidelity)
- Standardized **Modelfile** with SYSTEM prompt and stop tokens
- Integrations into **SlavkoPlatform™**, **SlavkoScore™**, **SlavkoAgent™**
- Ready for **Ollama Library** and **Hugging Face** publication

---

## 🚀 Installation

```bash
# Pull the kernel from Ollama Library
ollama pull Gertner/SlavkoKernel-v3

💻 Usage

# Run the kernel locally
ollama run Gertner/SlavkoKernel-v3

Recommendation: for reproducibility, always use the commit hash and local Modelfile.


🧬 Version Evolution

Version Focus
v1 Latency & Context Retention, basic SlavkoScore™ integration
v2 Adaptive Reasoning & Voice Handoff, branching & evaluation improvements
v3 Stabilization, replay fidelity, documentation, clear versioning

🧠 What v3 Brings

  • Replay Fidelity — prompt set for deterministic branching
  • Reproducibility — build and run repeatable on any machine
  • Naming Standard — clear versioning: v1, v2, v3
  • Documentation — README and CHANGELOG included
  • Platformization — SlavkoKernel™ as a modular orchestration layer, not just a model

🔗 Integrations

  • SlavkoScore™ — real-time evaluation
  • SlavkoAgent™ — multi-agent orchestration
  • Birtija Simulator — stress testing and chaos injection
  • SlavkoPlatform™ — meta-platform for modular AI infrastructure

🛠️ Modelfile (base v3)

FROM llama3.2

SYSTEM "You are a modular orchestration assistant with replay fidelity, milestone tracking, and scenario flag awareness."

PARAMETERS {
  "version": "Gertner/SlavkoKernel-v3",
  "commit": "REPLACE_WITH_COMMIT_HASH",
  "build_date": "REPLACE_WITH_ISO_DATE",
  "stop": ["<|start_header_id|>", "<|end_header_id|>", "<|eot_id|>"],
  "quant": "Q4_K_M"
}

TEMPLATE "<|start_header_id|>system<|end_header_id|> Cutting Knowledge Date: December 2023 {{ if .System }}{{ .System }}{{ end }}<|eot_id|>"

⚠️ License & AUP

  • License: LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
  • Acceptable Use Policy: Meta commitment to safe and fair use
  • You must read and comply with the AUP, especially for enterprise or commercial use.

🧭 Philosophy

SlavkoKernel™ v3 cements the foundation: reproducibility, documentation, and identity. Modular, transparent, and accessible to all developers, without unnecessary bloating.


🧪 Recommended Practices

  • Always use commit hash and local Modelfile for reproducibility
  • Run replay test suite for deterministic scenarios
  • Verify artifact hashes before distribution
  • Minimum hardware: RAM/VRAM requirements aligned with 3.21B Q4_K_M model

📜 References / Documentation

  • CHANGELOG.md — version history v1→v3
  • PROVENANCE.md — commit hash, build pipeline, source of weights
  • AUP_SUMMARY.md — Acceptable Use Policy summary

---

## **CHANGELOG.md** (ENG)

```markdown
# 📜 SlavkoKernel™ v3 CHANGELOG

All notable changes to this project will be documented here.

---

## [v3.0.0] — Stabilization & Identity (Latest)
**Release Date:** 2025-10-04

### Added
- Prompt set for deterministic replay fidelity
- Standardized Modelfile with SYSTEM prompt and stop tokens
- README, CHANGELOG, and PROVENANCE included
- Platformization features for orchestration layer

### Fixed / Improved
- Replay fidelity and branch determinism
- Documentation consistency
- Version identity clarity (v3)

### Notes
- Base model: LLaMA 3.2, 3.21B parameters, Q4_K_M quantization
- Ready for Ollama Library and Hugging Face publication

---

## [v2.0.0] — Adaptive Reasoning & Voice Handoff
**Release Date:** 2024-12-15

- Adaptive reasoning hooks for SlavkoAgent™
- Branching and evaluation logic improvements
- Initial experiments with Birtija Simulator

---

## [v1.0.0] — Latency & Retention Layer
**Release Date:** 2024-08-10

- Basic integration with SlavkoScore™
- Context retention layer for faster responses
- Basic replay tests introduced

PROVENANCE.md (ENG)

”`markdown

🧾 SlavkoKernel™ v3 Provenance

This document provides build and version provenance for reproducibility.


Model Base

  • Base: LLaMA 3.2
  • Parameters: 3.21B
  • Quantization: Q4_K_M (2.0GB)

Build Information

  • Version: Gertner/SlavkoKernel-v3
  • Commit Hash: REPLACE_WITH_COMMIT_HASH
  • Build Date: REPLACE_WITH_ISO_DATE
  • System Prompt: “You are a modular orchestration assistant with replay fidelity, milestone tracking, and scenario flag awareness.”
  • Stop Tokens: [“<|start_header_id|>”, “<|end_header_id|>”, “<|eot_id|>”]

Reproducibility Notes

  • All builds can be repeated using the same commit hash and Modelfile
  • Replay tests included in replay_tests/ folder ensure deterministic outputs
  • Hash verification recommended for all artifacts prior to distribution

Integrations

  • SlavkoScore™ — real-time evaluation
  • SlavkoAgent™ — multi-agent orchestration
  • Birtija Simulator — stress testing and chaos injection
  • SlavkoPlatform™ — modular AI infrastructure meta-platform

License & Use

  • License: LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
  • AUP Summary: Refer to AUP_SUMMARY.md