Start here

How the Kohnex engine works

Kohnex does not change your model. It changes how the model thinks — the engine injects expert reasoning into the model's chain-of-thought before it answers. The reasoning is inbuilt, and when nothing inbuilt fits, the engine forges new reasoning on the spot.

1

Ask the question

Your model gets the question from you — the same way it always does.

2

Engine forges the reasoning

forge() builds expert reasoning for your task — with a quality score from 0–100.

3

Check the score, tune it

Above 90: use it. Below: tune depth, rounds, population and retry until sharp.

4

Your model follows it

The same model now answers step-by-step — like a senior engineer, not a guesser.

Before (model alone):  "The bug is probably a typo. Check your code."

After (model + Kohnex engine): "1. Reproduce the failure reliably.
  2. Read the error message exactly.
  3. Trace the code path from input to failure.
  4. Check the assumptions the code makes.
  5. Fix the root cause, verify nothing else breaks."

Tutorial 01

Your first engine call

Give the agent a problem — forge() builds the reasoning, scores it, and your model follows the steps.

Our application is timing out under load. Use forge() to build
reasoning for diagnosing performance issues. Check quality.overall —
above 90, follow the steps exactly and keep the answer plain
and practical. Below 90, retry with depth="deep".
forge

Forge the reasoning

Describe the task. Returns expert steps plus a 0–100 quality score.

quality

Trust the score

Above 90 means battle-tested. Below means tune depth, rounds, population and retry.

follow

Follow the steps

Your model answers step-by-step — like a senior engineer, not a guesser.

Tutorial 02

Real-world examples

Three complete workflows — complex coding and debugging, security hardening, and product strategy. Copy each prompt as-is.

Example 1

Complex coding & debugging

A production service intermittently fails with a race condition. Watch how engine reasoning forces the model to think like a senior engineer — reproduce first, never guess.

Our checkout service intermittently fails with a race condition
under concurrent requests. Use forge() to build reasoning for
this kind of bug with depth="deep". Check quality.overall, then
follow the steps exactly: reproduce the failure, read
the error, trace the path, check the assumptions, and fix the
root cause. Show your reasoning at each step before the fix.
What the engine forces

No guessing. Every step must be shown before the fix.

Best modes

Pulse (diagnose), Spike (focus), Chain (prove it).

Why models need it

Models jump to "probably a typo." Engine reasoning forces structure.

Example 2

Security review & threat analysis

A security review is only as good as the thinking behind it. This prompt makes the model hunt like an attacker — including what it is not looking for.

Review this Node.js API for security vulnerabilities. Use forge()
to build reasoning for security review with depth="deep".
Check quality.overall, then follow the steps: list the attack
surface, trace each entry point, hunt for the least obvious
threat (what am I NOT looking for?), and rank findings by
exploitability. End with a practical hardening checklist.
What the engine forces

Attack-surface first, not a generic "use HTTPS" list.

Best modes

Shield (defense), Ghost (break it), Strike (adversarial).

Why models need it

Models list known issues; engine reasoning makes them hunt for the unknown.

Example 3

Product strategy & positioning

Strategic questions get a 4-step OODA-style treatment: observe the market, orient with evidence, decide between options, act with a plan.

We make a lightweight analytics tool and a big competitor just
dropped a free tier. Use forge() to build reasoning for
strategy with depth="standard". Check quality.overall, then
follow the steps: observe the market moves, orient with what is
actually changing, weigh our options (fight, pivot, focus,
partner), and recommend one decision with a 90-day action plan.
Challenge your own recommendation at the end.
What the engine forces

Observe → orient → decide → act. No gut-reaction strategy.

Best modes

Warpath (strategy), Radar (market scan), Fork (futures).

Why models need it

Models answer "lower your price." Engine reasoning forces a full strategic loop.

Quick reference

Forge playbook: which settings for which task?

One tool, tuned per task. Describe the job, check the score, tune until sharp:

Task Forge call
Fix a bugforge, depth="deep"
Security reviewforge, depth="deep"
Architecture callforge, depth="standard"
Quick questionforge, depth="quick"
High-stakes buildforge, rounds=3, population=5
Niche domainforge, domain="..."

Model Guides

Make your model smart

The same engine flow works on any model — frontier or open-weight. Here's how it plays out.

GPT + Kohnex for Code Review

Ask for a review with the engine finding security reasoning. GPT checks each file line-by-line instead of giving a generic "looks good."

"Review this PR. Use forge() to build reasoning for code review with depth="deep", check quality.overall, and follow the steps exactly."

DeepSeek + Kohnex for Debugging

DeepSeek gets diagnosis reasoning from the engine: reproduce → read error → trace path → check assumptions → fix root cause.

"Debug this failure. Use forge() to build reasoning for debugging, check quality.overall, follow the steps exactly."

Claude + Kohnex for Architecture

Architecture questions get systems reasoning: parts, connections, feedback loops, leverage points — not just a diagram.

"Design the architecture for X. Use forge() to build reasoning for architecture, check quality.overall, follow the steps."

Llama + Kohnex for Strategy

Llama uses decision reasoning to weigh options with evidence and commit — observe, orient, decide, act.

"Decide between X and Y. Use forge() to build reasoning for decisions, check quality.overall, follow the steps."

Go deeper

Compose the engine

A strong workflow describes, forges, checks, and tunes — the engine covers the whole arc.

1. forge()            → engine builds the reasoning + quality score
2. Check score         → above 90: use it; below: tune and retry
3. Tune                → depth, rounds, population up until sharp
4. Cite the pattern    → name it in your answer
5. Follow the reasoning → apply the steps to your question