Unikernal-v9

Start Doctor
Start Kernel
Chat Demo
Code Agent Demo
TTS Demo

🧠 Unikernal v9.1 — FreeStack Polyglot AI Operating System 🌟 Inspiration

Modern AI projects waste enormous time wiring together LLMs, voice systems, Python tools, APIs, and infrastructure. Every hackathon, every startup, every team ends up rebuilding the same integrations over and over again.

I asked myself:

“Why doesn’t AI have an operating system?”

Not a model. Not a framework. A kernel — that makes Python, Node, LLMs, voice, APIs, and workflows all talk through a single universal layer.

This became the inspiration for Unikernal v9.1: A FreeStack AI OS that works entirely on free-tier providers (Groq, ElevenLabs, HuggingFace optional), so it is accessible, fast, and hackathon-friendly.

⚙️ What it does

Unikernal v9.1 is a universal interoperability engine that gives AI applications a single kernel for:

✔ Ultra-fast LLM generation (Groq) ✔ High-quality text-to-speech and speech-to-text (ElevenLabs) ✔ Code generation and specialized coding agents ✔ Polyglot tool execution (Python / Node.js / Shell) ✔ Workflow automation via YAML flows ✔ Distributed mesh networking ✔ Fault tolerance, retries, and a Dead Letter Queue ✔ Zero-trust security with scope-based access ✔ Hot-reloading tools and flows without restarting

In short:

Unikernal lets any language talk to any AI or tool, instantly, through one kernel.

Developers can build chatbots, voice assistants, IoT controllers, ETL pipelines, or multi-agent systems without rewriting integrations.

🛠️ How we built it

Unikernal v9.1 is built as a modular AI OS with several layers:

Universal Messaging Layer (UDL/UDM)

UDL = Universal Data Language (call/event/stream messages)

UDM = Universal Data Model (tools, services, agents)

This standardizes how every tool and agent communicates.

Routing Kernel

Central brain that routes all messages to:

Groq LLM Adapter

ElevenLabs TTS/STT Adapter

Python Tool Adapter

Node Tool Adapter

Shell Tool Adapter

HTTP Tool Adapter

It supports:

Idempotency

Retries with exponential backoff

Dead Letter Queue

Mesh forwarding over HTTP

FreeStack AI Layer

Groq → primary LLM provider (70B/8B models, ultra-fast)

ElevenLabs → Voice generation (new free-tier models)

HuggingFace (optional) → backup LLM provider

Configured via:

config/llm.runtime.json .env

Agent System

Support Agent

Code Agent

Voice Agent

Orchestrator Agent

Each built on YAML flows and tool definitions.

Tooling Layer

Custom tools:

python:sum_numbers

node:format_text

http:get_joke

shell:list_files

ETL fileSource / fileSink tools

Diagnostics Layer

A built-in Doctor:

npm run doctor

Checks:

Node version

Ports

.env validity

Provider connectivity

Config validity

Runtime Layer

Hot Reloading (config/tools.yaml & flows)

Mesh nodes (config/mesh.nodes.yaml)

Structured logs

🧩 Challenges we ran into ❌ Model deprecations

Groq and ElevenLabs deprecated old free-tier models, requiring dynamic model routing and config overrides.

❌ HuggingFace API migration

Old inference API (api-inference.huggingface.co) was shut down → upgraded to router-based endpoints.

❌ Body parsing mismatch for TTS endpoint

Voice API needed multi-format parsers for text/plain, json, and form-data.

❌ Distributed mesh networking reliability

Solved using:

HTTP forwarding

Node registry

Heartbeats (future work)

DLQ for message failures

❌ Hot-reloading correctness

Updating flows/tools while running required careful cache invalidation.

🏆 Accomplishments we’re proud of ⭐ Built a complete AI Operating System from scratch ⭐ Achieved full polyglot execution (Python + Node + Shell + HTTP) ⭐ Integrated LLMs and TTS on fully free-tier APIs ⭐ Designed a universal data model (UDL/UDM) ⭐ Implemented fault tolerance, retries & DLQ ⭐ Built mesh networking between kernel nodes ⭐ Voice pipeline works end-to-end (JSON → MP3 output) ⭐ Created a dev experience with:

npm run dev

npm run doctor

npm run test:api

⭐ It runs locally, with no cloud cost

Perfect for hackathons.

📚 What we learned

How to build a kernel-style architecture for AI systems

How to design universal agent/tool interfaces

How to integrate multiple LLM providers gracefully

How to handle model deprecations dynamically

How to build robust voice pipelines with ElevenLabs

How to design ETL workflows using YAML

How to use observability patterns (structured logs, DLQ, retries)

How to design AI systems that run offline / on free-tier infrastructure

🚀 What’s next for Unikernal-v9 🔜 v9.2 — Vision Agent (image understanding on Groq) 🔜 v9.3 — Memory Engine (long-term project memory) 🔜 v9.4 — Real-time WebSocket Streaming 🔜 v10 — Portable Edge Edition for IoT devices 🔜 Unikernal Hub — tool marketplace (community plugins) 🔜 Android / iOS client SDK 🔜 Web dashboard UI (flows, logs, testing tools) 🔜 Multi-agent orchestrator with auto-tools

And ultimately:

Turn Unikernal into a universal AI OS used across industries and hackathons worldwide.