Model Interface & Customer Layer for Efficient Unified Semantics
SEMA — Semantic Engine for Multi-model Activation
Local
Ollama
qwen2.5 · llama · mistral
API
OpenAI
GPT-4o · GPT-4 mini
API
Anthropic
Claude Sonnet · Haiku
Free
Groq
Llama 3 · Mixtral · Gemma
Model: Ollama local
SEMA active. NICLEUS is filtering your input before it reaches the model.
0
Tokens in
0
Tokens out
—
Out / In ratio
These are the real tokens exchanged with the model in this session.

NICLEUS shapes every prompt before it reaches the model — tone, depth, semantic lens. Extensively tested on standard hardware and across providers: token reduction ranges from 8.8% on short operational queries up to 90% on complex ones. Anti-jailbreak interception: 100%. Full benchmark: github.com/aleinTour/nicleus

Send the same question to several models at once. NICLEUS prepares the prompt identically for all — the Vailati Filter scores the responses on signal density, honesty, and coherence. The cleanest answer wins.

Using local Ollama: start it with OLLAMA_ORIGINS=* ollama serve so your browser can reach it. Your model, your RAM, your call.

NICLEUS compresses and routes every input automatically — no commands needed. Savings happen by default.

Modus is the next layer. A set of single-letter keys that shape how NICLEUS thinks before answering — tuning tone, depth, memory, and data sources. Not a requirement. A precision instrument for those who want more control.

SOMA loop commands
a
Intuitive SOMA
Free, associative response. Ironic or satirical when it fits. High creativity, low filter.
b
Brief SOMA
Short answer, no memory written to buffer. Fast, disposable, no context trail.
e
Analytical SOMA
Structured numbered response. 1. 2. 3. — each point on its own line. Precise, methodical.
w
Web SOMA · main
Fetches live data before answering. Weather via wttr.in, facts via DuckDuckGo Instant Answer.
s
Save SOMA
Saves the current exchange to the local JSON dataset. Manual — never automatic.
r
Deep research SOMA · main
Long-form response, scientific style. Extended token cap, extended timeout. For complex questions.
q
Quit SOMA · main
Exit the current loop. In main: saves session stats to JSON before closing.
Main loop flags
-w
Web flag main
Append to any query to activate web layer. Example: meteo roma -w
-r
Free response main
Removes token cap. Use for long technical questions that get cut short.
!file
File context main
Injects a local file into the prompt. Supports .rs .txt .toml — example: !main.rs
↩
Continua main
Type "continua" to activate context buffer. Max 3 exchanges — then NICLEUS educates you on how to proceed.
soma
Enter SOMA main
Launches the SOMA/VAILAT symbiont loop. Give it a seed — it generates scenarios and iterates.

A plain-language guide for people who are not developers. No jargon, no setup headaches.

What MICLEUS actually does

MICLEUS lets you talk to an AI model — the same technology behind ChatGPT or Claude — with a filter called NICLEUS in front of it. NICLEUS reads what you write, understands what you mean, and sends the model a clean, focused version of your question instead of your raw words.

Faster answers. Lower cost. Nothing stored anywhere.

Connect in five steps
1
Pick a model
On the Connect tab, choose Anthropic, OpenAI, Gemini, Groq — or your own local Ollama.
2
Paste your API key
The field appears once you pick a provider. Don't have a key? See below.
3
Click Connect
You land straight in the Chat tab.
4
Write normally
Plain language, your own words. No prompt engineering required.
5
Use Focus mode
The ⛶ button expands the chat for real work.
Your key is never saved
It lives in your browser's memory for this session only. Close the tab and it's gone. It goes nowhere except to the provider you chose.
Where to get an API key

Google Gemini — easiest start, generous free tier.
aistudio.google.com → Get API key → Create API key (key starts with AIza…)

Anthropic (Claude) — needs a small credit top-up.
console.anthropic.com → API Keys → Create Key (key starts with sk-ant-…)

Groq — very fast, free tier available.
console.groq.com → API Keys → Create Key (key starts with gsk_…)

For businesses — monadini & private corpus

NICLEUS can be extended with monadini: small self-contained modules that answer specific question types instantly, without touching the AI model at all. Already built: weather, financial markets (crypto, gold, forex), and a closed-corpus identity module.

That last one is the business case. Load your company's own documents, procedures, or policies — employees and customers ask in plain language, and the system answers only from verified material. No guessing, no invented answers, and a fraction of the token cost.

Custom monadini, private knowledge bases and dedicated deployment are available on request — see the Contact tab.

Then try

The Modus buttons under the chat box (Deep, Structured, Quick) to change how answers are shaped · Agorà to compare several models on the same question · the 📚 corpus icon to load your own knowledge base.

The internet must serve humanity.

MICLEUS is the mind. NICLEUS is the heart. Together they form SEMA — a semantic gateway that sits between you and the AI model of your choice, reducing consumption, protecting your identity, and amplifying response quality.

For individuals it's free. Bring your own API key or connect Ollama locally. Zero data collected, zero profiling.

For businesses we offer custom integration, proprietary datasets generated by SOMA, and a semantic security layer built around your needs.

SEMA
Semantic Engine for Multi-model Activation — a new scientific category. Not a chatbot. Not a wrapper. A Rust-powered cognitive routing layer that processes language before it reaches any model.

berylabs.com · 2026