Cognitive Stack

Your AI doesn't just generate. It reasons.

Same question. Same model. Completely different answer.

The Cognitive Stack gives your AI structured reasoning, modeled after real biological cognitive systems, that runs behind the scenes before it speaks.

You see the result. Not the process.

Without Cognitive Stack
" Should I use mycorrhizal fungi in my organic grow?"
"Great question! Mycorrhizal fungi are amazing for plants. They can improve nutrient uptake by up to 40% and are widely recommended for organic grows. You should definitely consider adding them to your soil..."
No sourcesNo confidence scoreNo caveatsGeneric
With Cognitive Stack
"Should I use mycorrhizal fungi in my organic grow?"
Yes, but with caveats. Arbuscular mycorrhizal fungi can meaningfully improve phosphorus uptake in organic soil. However, the widely cited 40% yield improvement has no peer-reviewed basis, and effectiveness depends heavily on your soil pH.
Confidence: 0.74Moderate: some claims unverified
Evidence
82%
Reasoning
74%
Calibration
61%
Supported: Arbuscular mycorrhizal fungi improve phosphorus uptake in organic soil. 3 peer-reviewed studies via PubMed.
Challenged: The "40% improvement" figure has no peer-reviewed basis. Likely drawn from marketing materials, not clinical data.
Blind spot detected: pH dependency. Below 5.5, mycorrhizal colonization drops sharply. Your soil pH matters more than the fungi species.
Self-corrected: Initial confidence was 0.91. Recalibrated after finding the 40% figure in 0 of 18 databases checked.

This is what your AI's output looks like after the Cognitive Stack processes it. The reasoning happens invisibly, inside the API call. You keep using ChatGPT, Claude, or whatever you use now. The answers get better.

Try it free$9/mo · Works with any AI provider · Cancel anytime
Works with
Extension
ChatGPT
Gemini
Claude Desktop
API
CLI
Getting started

Three steps. Five minutes.

1
Get your AI key
From OpenAI, Anthropic, Google, Mistral, DeepSeek, whoever you use today.
2
Add it to ASURIQ
Paste it once. We never store it. It calls your provider directly.
3
Your AI thinks now
Every response comes back with confidence, challenges, and blind spot detection.

Don't have an AI key? The Chrome extension and custom GPT work without one for free features.
Need help getting a key? Step-by-step guides for every provider →

11 cognitive systems · modeled after biological architecture · working simultaneously on every response
The architecture

Modeled after biological cognition.

Your brain doesn't have one thinking mode. It has systems. Curiosity that pulls you toward gaps in what you know. An immune system that rejects bad information. Sleep that consolidates memories. Adversarial challenge that stress-tests beliefs before you commit.

We built all eleven.

Seven of these are reasoning tools you call directly through the API: confidence scoring, adversarial challenge, calibration, reflection, growth detection, and assertion routing. The other four (LOOM, brain.db, NIGHTSHIFT, TESSRYX) are infrastructure. They run continuously underneath, maintaining memory, consolidating knowledge, tracking provenance. You don't call them. They make everything else work.

Baars: Global Workspace TheoryACT-R: Memory Decay ModelHormesis: Stress-StrengtheningAutopoiesis: Self-Creating Systems
LOOMGlobal Workspace
All reasoning frameworks active at once. Depth through parallax, not sequential analysis
LANTERNCuriosity Drive
Finds gaps in what your AI knows and investigates them autonomously
PROMETHEUSAutopoiesis
Spots what your AI can't do yet and proposes ways to grow
WHETSTONEHormesis
Stress that strengthens. Generates the strongest possible counter-argument
IMPRINTNeuroplasticity
Reviews what happened and feeds learning into the next session
TREGImmune System
Catches overconfidence and institutional bias in training data
brain.dbHippocampal Memory
Persistent memory that strengthens useful connections, weakens unused ones
NIGHTSHIFTSleep Consolidation
13 autonomous passes that prune, synthesize, and detect drift. Runs while you sleep
TESSRYXConnectome
Tracks where every piece of knowledge came from and what depends on it
YUMAQuality Gating
Validates output quality before it reaches you
AUTONOMICMotor System
Executes tasks without supervision, with self-healing when things fail

They're multiplicative. WHETSTONE can't challenge effectively without TREG calibrating what "effectively" means. IMPRINT can't learn without brain.db to store what it learned. NIGHTSHIFT can't maintain memory it can't access. Remove any one system and the others degrade. The whole is exponentially greater than the sum.

"Can't I just write a better prompt?"

What a prompt does
"Be critical of your answer"
Same model · Same weights · Same training data
"I've carefully considered alternatives..."
(agrees with itself)
What the Cognitive Stack does
Confidence: 7-axis quality scoring
WHETSTONE: separate adversarial engine
TREG: calibrated against 15 failure cases
IMPRINT: learning fed back into next session
Answer: 4+ cognitive processes deep

The difference between a prompt and an architecture is the difference between asking someone to check their own homework and hiring an auditor. Your AI will always agree with itself. The Cognitive Stack runs genuinely independent systems with different instructions, different calibration data, and different failure modes.

7 reasoning tools · 4 infrastructure systems · calibrated against 15 documented failure cases
Two modes. One architecture.

One model thinks deeper. Multiple models think wider.

Single-model mode

Same model you use today. Deeper reasoning. Your existing AI gains structured thinking: confidence scoring, adversarial challenge, calibration, reflection, growth detection. All 7 tools running on your one provider.

You need
One AI provider key (Claude, GPT, Gemini, etc.)
Cognitive Stack subscription ($9/mo)
That's it
$9/mo
Included with Cognitive Stack

Multi-model mode

Same 7 tools, but running across multiple providers. Cross-provider confidence scoring. WHETSTONE challenges from a genuinely different model. Calibration that catches provider-specific blind spots. Independent errors, compounding accuracy.

You need
Multiple provider keys (Claude + GPT, etc.)
Cognitive Stack subscription ($9/mo)
Routing add-on ($13/mo), manages your provider keys
$19/mo
Intelligence Layer bundle: save $3/mo vs separate

Start single-model. It's where most users begin. When you want cross-provider validation, where Claude challenges GPT's confidence, or DeepSeek catches blind spots that Gemini misses, add Routing and unlock multi-model mode. Note: some tools (Confidence, WHETSTONE, IMPRINT, ARGUS) require a mid-tier model. Without Routing, they won't fire on budget models. ASURIQ won't badge degraded output.

Start free. Add power when you need it.

Free forever
Verification badge on AI answers
Basic confidence scoring
Systems Map (all 11 systems)
Chrome extension
No API key required
No credit card
No time limit
Cognitive Stack · $9/mo
Everything free, plus:
Full 7-axis confidence analysis
WHETSTONE adversarial challenge
TREG epistemic calibration
IMPRINT reflection + learning
PROMETHEUS gap detection
Assertion Router
Continuous framework updates
Uses your existing AI provider key. Cancel anytime.
Single-model reasoning

Seven reasoning tools. Four infrastructure systems.

The seven reasoning tools are what you call. Each makes your AI smarter on its own. Together, they compound: confidence scoring means nothing without calibration, and calibration means nothing without adversarial challenge. Some tools require a mid-tier model or better. If your model is below the threshold, the tool explains why and offers Routing as the fix.

Free
Systems MapThe table of contents

See every cognitive system and how they connect. The architecture at a glance. Always free, no key needed.

Confidence Score7-axis quality assessment

Not a single number. Seven dimensions. Evidence quality, reasoning depth, calibration, source reliability, domain expertise, internal coherence, and meta-awareness. Shows you exactly WHERE the answer is strong and where it's weak.

Assertion RouterRight mode, right response

Some questions deserve confident assertion. Some need investigation. Some need honest refusal. Your AI usually answers everything the same way. The router picks the mode that matches the question.

WHETSTONEAdversarial challenge

Not "here's the other side." The strongest counter-argument a genuinely different mind can construct. Like stress-testing steel. What survives is stronger for having been tested.

TREGEpistemic immune system

Is that 80% confidence actually warranted? Calibrated against 15 documented cases of epistemic failure across five categories. The immune system that catches institutional bias in training data.

IMPRINTReflection + learning

After a session, your AI reviews what happened. What surprised it, what it should update, what tools would have helped. Those learnings feed directly into the next session. Neuroplasticity, implemented.

PROMETHEUSGrowth detection

Most AI just answers your questions. PROMETHEUS spots the questions it SHOULD be asking but isn't. Detects capability gaps, proposes upgrades. Your AI doesn't just work. It gets better.

Multi-model Cognitive Stack

Same tools. Different brains. Better answers.

Every cognitive tool in the stack works with one model. But when you add Routing and connect multiple providers, those same tools gain a superpower: cross-provider validation. Your confidence score isn't one model's self-assessment anymore. It's three independent models evaluating the same claim.

Confidence
Single-model: One model scores its own certainty
Multi-model: Three models score the same claim independently. Disagreement between scores reveals the questions where confidence is least trustworthy.
WHETSTONE
Single-model: Same model generates counter-arguments
Multi-model: A genuinely different model generates the challenge. Different training data means different blind spots. The counter-argument comes from a mind that learned different things.
TREG
Single-model: Calibration against one model's failure modes
Multi-model: Cross-calibration across providers. Claude's overconfidence profile is different from GPT's. TREG learns which provider to trust in which domain.

Requires Routing. Multi-model Cognitive Stack needs Routing ($13/mo) to manage your provider keys and route queries across models. Learn about Routing →

What this replaces
Peer review / fact-checking services$50–200/mo
Multiple reasoning/critique tools$15–30/mo
$65–230/mo replaced by $9/mo
Your data stays yours
Prompts stay with your provider. We see analysis metadata only.
Keys never stored
One-way hash for authentication. Your credentials pass through. Never persist.
~2 second responses
Single-model cognitive tools return in about 2 seconds.
37 of 100 flagged
We ran 100 ChatGPT answers through verification. See the study →

Currently in friends-and-family beta. Built on peer-reviewed cognitive architecture.

Baars — Global Workspace TheoryACT-R — Memory Decay ModelWang et al. 2025 — Silent AgreementLi et al. EMNLP 2024 — Sparse Debate

Make your AI think.

$9/mo · Works with any AI provider · Cancel anytime