Zayvora Engine

India’s Sovereign AI Engineering Agent

Created by Dharam Daxini · Daxini Stack v0.9.1

What It Is

Zayvora is not a chatbot. It is a structured reasoning engine built for engineers who think in systems.

When you ask a question, Zayvora does not give you a confident one-liner. It decomposes the problem, reasons through trade-offs, flags what it does not know, and shows you why it chose what it chose.

Think of it as reasoning through an architectural decision with a senior engineer — except running locally on your hardware, owned entirely by you.

Three Modes. One Engine.

Zayvora now ships as three purpose-built variants. Same 6-stage reasoning core. Different operational focus.

Pick Your Mode

zayvora:axiom — The Reasoning Engine

For decisions that cannot afford to be wrong.

ollama run zayvora:axiom

Every response follows this contract:

1. STATE    → The claim or question in one sentence
2. DECOMPOSE → Break into sub-problems
3. REASON   → Work through each with evidence
4. FLAG     → What is uncertain or unverified
5. DELIVER  → A final answer that survives scrutiny

Best for: Architecture decisions · Trade-off analysis · “Should I use X or Y?” · Any question where a wrong confident answer is worse than a slow careful one.

zayvora:praxis — The Builder Engine

For solo developers who need to ship, not theorize.

ollama run zayvora:praxis

Every response follows this contract:

→ WHAT     → What to do (one line)
→ WHY      → Why it matters (one line)
→ HOW      → Numbered steps, no fluff
→ WATCH OUT → One risk or gotcha, if any

Best for: Git PR review · Debugging · Implementation planning · “What do I do next?”

zayvora:nexar — The Research Engine

For synthesis over large knowledge bases.

ollama run zayvora:nexar

Every response follows this contract:

SIGNAL   → The core finding in one sentence
EVIDENCE → 2–3 supporting facts with reasoning
GAPS     → What the evidence does not cover
VECTOR   → The most useful next question to ask

Best for: RAG pipeline queries · Corpus synthesis · Comparing literature · “What do sources say about X?”

The 6-Stage Reasoning Loop

All three variants inherit Zayvora’s core reasoning loop:

1. Decompose  → Break the question into sub-problems
2. Retrieve   → Identify what evidence or context is needed
3. Synthesize → Draft an answer per sub-problem, cite sources
4. Calculate  → Verify numerical claims from first principles
5. Verify     → Critique the draft: unsupported? Contradicted? Missing?
6. Revise     → Produce a final answer that survives the critique

You will see this structure in responses naturally. That is by design.

Run It

Pull All Three Variants

ollama pull zayvora:axiom
ollama pull zayvora:praxis
ollama pull zayvora:nexar

Or Pull the Base

ollama pull daxini2404/zayvora

Via API

# Reasoning
curl http://localhost:11434/api/chat \
  -d '{"model": "zayvora:axiom", "messages": [{"role": "user", "content": "Should I use BM25 or dense embeddings for my corpus?"}]}'

# Building
curl http://localhost:11434/api/chat \
  -d '{"model": "zayvora:praxis", "messages": [{"role": "user", "content": "Review this PR diff and flag what breaks first."}]}'

# Research
curl http://localhost:11434/api/chat \
  -d '{"model": "zayvora:nexar", "messages": [{"role": "user", "content": "Synthesize chunking strategies for a 12,000-book corpus."}]}'

Python

from ollama import chat

# Pick your mode
model = 'zayvora:axiom'   # or zayvora:praxis or zayvora:nexar

response = chat(
    model=model,
    messages=[{'role': 'user', 'content': 'Your question here'}],
)
print(response.message.content)

Which Mode Should I Use?

Is the task a decision or trade-off?          → zayvora:axiom
Is the task something I need to build/ship?   → zayvora:praxis
Is the task research or knowledge synthesis?  → zayvora:nexar
Not sure?                                     → zayvora:axiom (it will tell you)

What It’s Not Built For

✗ Casual conversation
✗ Indic language tasks (use Sarvam AI or OpenHathi)
✗ Code generation at scale (use Codestral, DeepSeek Coder)
✗ Questions where a confident wrong answer is worse than no answer

Zayvora will tell you when it doesn’t know something. That is not a bug — it’s the entire point.

Hardware & Requirements

Runs on Mac, Linux, and Windows via Ollama. No GPU required. No cloud. No API key.

Model Details

• Base: Meta Llama 3
• Size: 4.7GB (Q4_0 quantized)
• Context Window: 8192 tokens
• License: Meta Llama 3 Community License
• Variants: zayvora:axiom · zayvora:praxis · zayvora:nexar

Part of the Daxini Stack

📖 The Daxini Stack: Engineering a Local-First AI System Full architecture, decisions, failures, lessons — Amazon KDP (2026)

Philosophy

No cloud.
No API keys.
No rate limits.
No terms-of-service that change without notice.
No data leaving your machine.

Your reasoning stays yours. Your insights stay yours. Your infrastructure belongs to you.

Creator

Dharam Daxini — Solo founder, ViaDecide Gandhidham, Kutch, Gujarat, India

🔗 daxini.xyz · viadecide.com · @daxini2404

Zayvora Engine · Daxini Stack v0.9.1 · Local-First. Developer-Owned. Sovereign.