A local-first AI agent that grows its own capabilities, discovers new skills, writes code, tests itself, and evolves, all on your own hardware.
3D Agent Avatar
A command or query triggers the pipeline.
Analyzes if the request exceeds current capabilities.
Search ecosystem and local repositories.
Code-generation of new skills when no match is found.
Ephemeral replica reviews code for quality and safety.
Human confirmation gates the resolution.
Skill is merged into the ecosystem.
A command or query triggers the pipeline.
Analyzes if the request exceeds current capabilities.
Search ecosystem and local repositories.
Code-generation of new skills when no match is found.
Ephemeral replica reviews code for quality and safety.
Human confirmation gates the resolution.
Skill is merged into the ecosystem.
The system is designed around a modular, tool-first philosophy. Every component is replaceable, every skill is independently testable, and the whole stack runs on consumer hardware.
The model always has access to tools. Skills and routines are meta-tools in a native tool-calling loop. No routing-first triage.
ADR-022Multi-slot architecture: primary (Nemotron), audio (Gemma 4), draft, embed — each independently managed with LRU eviction.
ADR-018GPU-accelerated model serving with Nemotron-3B (64k ctx) for local inference. Gemma 4 for audio. Ollama drop-in for cloud models.
ADR-013Ephemeral agent instances (critic, planner, pipeline roles) with isolated system prompts and named slot routing.
ADR-008Idle reflection: System 1/2 gap detection, identity and memory consolidation, proactive Telegram thoughts.
ADR-005Evolution anchored to real user failures. Only fires when a request genuinely fails. Human confirmation gates resolution.
ADR-020Auto-discovers OpenClaw, coding agents (opencode, Aider, Claude Code), and peers via port scan, mDNS, and MCP.
ADR-015Three-layer composable context: token-aware history, embedding retrieval, and skill-driven context providers built fresh per inference.
ADR-014Kernel Evolving runs entirely on your own hardware. Everything is local-first with cloud escalation only for code synthesis. No data leaves your machine unless you explicitly allow it.
Requirements Python 3.11+, CUDA GPU (8GB+ VRAM)
Every skill is autonomously generated, tested, and curated by the agent. Each one lives in its own git branch with a full SKILL.md spec, command interface, and test suite. Skills are synced from kernel-synthesized-skills .
Drafts a 3-paragraph blog introduction about why agent briefs matter, then critiques it for clarity.
Set up a cron job to compress log files older than 7 days and remove very old archives
Analyze the last 24 hours of kernel evolution activity and write a concise self-assessment report
Generate a weekly performance report from server logs and email it
Send a proactive Telegram message using a bot token and chat id from environment variables.
Schedule Telegram reminder messages by generating a one-shot send script and local scheduler instructions
Translate a document from Italian to English while preserving basic formatting and document structure
Translate the provided Italian text to English and save it to ~/kernel-evo-notes/translation.txt
Monitor a website for content changes and send Telegram notifications when updates are detected.
Resize a batch of product images and apply a text watermark for e-commerce use.
Convert all images in a folder to optimized WebP format with configurable quality and recursion.
Sync local markdown notes to a Notion database via the Notion API
Fetch the Wikipedia GDP (nominal) page, extract the first 5 rows from the main country GDP table, and save them as JSON.
Extract structured data from an HTML table and save it as JSON.
Convert Markdown files into PDF documents when a user needs shareable, printable, or archival output from Markdown content.
Build a CLI tool that queries a PostgreSQL database and exports CSV
Detect anomalies in a time-series CSV dataset and print alerts for outlier points
Transcribe a meeting audio or video recording with ffmpeg Whisper and extract action items from the transcript.
Deploy a static site from a local build folder to GitHub Pages
Parse an OpenAPI spec and generate a typed Python client package from it
Review a Python function for bugs and improvements, then return a corrected version with inline comments explaining each change.
Generate Python unit tests for a module using static analysis of its source code.
Generate pytest unit tests for a Python module containing calculate_roi(invested, returned) -> float
Generate unit test skeletons for a Python module using static analysis of source code.
Fetch an article from a URL and return 3 key points as a concise summary
Fetch a web article and return its 3 key points as concise bullets.
Fetch a web article and return 3 concise key points.
Fetch a YouTube video transcript and produce a concise summary
Summarise a YouTube video transcript into concise bullet points
Search online for AI Agents information and fetch the latest arXiv paper matching the topic.
Search arXiv for papers about agentic AI and return a concise structured summary.
Find a recent arXiv paper about self-evolving AI agents and return the title and abstract
Find a recent arXiv paper about self-evolving AI agents and return the title and abstract
Handles generic "test" tasks by echoing input and returning a successful result