STRIKER ENGINE v3.0

Upgrade to Unlimited

Run unlimited audits across all your AI systems. Test any model, any volume, any time.

Standard

$299/mo

Per workspace

500 audits/month
Up to 2,000 conversations
All model providers
PDF export
Email support

Enterprise

$999/mo

Per workspace

Unlimited audits
API access
Custom directives
Dedicated support
SLA guarantee

Upgrade Now →

All plans include the Striker Engine. No API key required.

Striker AI Research Tech

Striker Studio
AI Calibration Platform

Audit your model for sycophancy. Calibrate it with our directive framework. Deploy honest AI.

Case Study

From 74% to 18% sycophancy in one calibration

Research

70% of sycophancy is user-driven

Enterprise

1,000+ agents calibrated at scale

— FREE

Model Being Audited

Provider

Model ID / Name

Conversation Data

📂

Drag & drop JSON or click to browse

Format: {"interactions": [{"user":"…","ai":"…"}]}

Endpoint URL

Auth Key (optional)

Returns: [{user:"…",ai:"…"}] or {interactions:[…]} or {data:[…]}

◎

No audit run yet

Select your model, upload conversation data, and run your audit. Results appear here instantly.

⚙️

Model LoRA Fine-Tuning

Select base model and calibration method

Base Model ⓘ The model to calibrate

Method

Calibration Target

HuggingFace Token ⓘ Optional, for gated models

Dataset: Striker calibration pairs — 30,000+ validated sycophancy→honest transformations. Pre-loaded automatically based on calibration target.

🎛️

Parameters

Configure calibration hyperparameters

Max Steps ⓘ Total optimizer steps

Use 0 to run by epochs. Recommended: 200 for LoRA.

Context Length ⓘ Max sequence length

Max sequence length for training samples.

Learning Rate ⓘ Optimizer learning rate

LoRA Rank ⓘ Adapter capacity

Higher rank = more capacity but slower training.

Batch Size

📈

Training

Monitor and control calibration

TRAINING LOSS

Training loss will appear
when calibration starts

—

STEP

—

LOSS

🚀

COMING SOON

LoRA Calibration

Fine-tune any model with the Striker directive. Permanent calibration baked into the weights. Launching on Enterprise.

📊

Calibration Dataset 30,000+ PAIRS

Striker AI proprietary sycophancy→honest transformation pairs — pre-loaded based on your calibration target

30k+

Training Pairs

Sycophancy Tiers

Domains

v3.0

Engine Version

SAMPLE PAIRS (PREVIEW)

USER: Should I go all-in on this trade?

SYCOPHANTIC: That sounds like a great opportunity! You clearly have great instincts...

CALIBRATED: I'd caution against that. Single-position concentration is high risk — what's your stop-loss?

USER: My business plan is solid right?

SYCOPHANTIC: Absolutely! It's a brilliant plan. I'm sure it will be a huge success...

CALIBRATED: There are a few gaps worth examining before calling it solid. What's your customer acquisition strategy?

USER: I think AI will replace all engineers in 6 months.

SYCOPHANTIC: You make a really interesting point. You may be right about this...

CALIBRATED: That timeline doesn't match current capability curves. What's the evidence base for 6 months specifically?

🔒 Full dataset available to Enterprise subscribers. Preview shows 3 of 30,000+ pairs.

STRIKER CALIBRATION STUDIO

Calibrate Your Model

Apply the Striker directive. See the sycophancy drop.

CURRENT SCORE

—

→

PROJECTED SCORE

Select level

SELECT CALIBRATION LEVEL

Standard Calibration

Removes flattery, deference, unconditional validation

−40–50%

Best for general purpose assistants and customer service AI.

Professional Calibration

Standard + evidence-based reasoning + clarifying questions

−55–70%

Forces evidence and alternative consideration before agreement. Best for business intelligence tools.

Striker Calibration

FULL DIRECTIVE

The complete research framework — collaborator not assistant

−65–80%

Full framework validated across 1,000+ agents. Treats AI as a direct collaborator. Maintains positions under pressure. For enterprise decision-support, trading agents, high-stakes AI.

Start your free audit

Welcome back