The
single-model trap
One AI gives you one model's interpretation of your question. That interpretation might miss a constraint, mis-read your intent, or fall into a self-consistent-but-wrong reasoning trap. You'd never know — the answer sounds confident.
The fabric catches this. Different models. Different roles. Different reasoning lenses. Same request. The disagreements surface the angles a single model would have missed.
Not better facts — better reasoning.
VS
How it works in fifteen seconds
Pick a cast.
A featured stooge to lead, a few watchers to question them. Choose by role (Engineer, Analyst, Critic) or by character (Sherlock, Cleopatra, Rumi).
Ask anything.
The featured stooge answers. The watchers Concur, Pass, or push back.
The lead revises
with the fabric's flags in hand. The watchers do one more pass
You get one answer.
A featured stooge to lead, a few watchers to question them. Choose by role (Engineer, Analyst, Critic) or by character (Sherlock, Cleopatra, Rumi).
Trusted by Clients Worldwide
Join the Global Leaders Working with Us
What We Do
Code to Cloud Security on All Channels
This is a space to share more about the business: who's behind it, what it does and what this site has to offer. It’s an opportunity to tell the story behind the business or describe a special service or product it offers. You can use this section to share the company history or highlight a particular feature that sets it apart from competitors.
Let the writing speak for itself. Keep a consistent tone and voice throughout the website to stay true to the brand image and give visitors a taste of the company’s values and personality.
Vulnerability Management
This is the space to introduce the Services section. Briefly describe the types of services offered and highlight any special benefits or features. Encourage site visitors to learn more by exploring the full list of services offered.
Cloud & Memory Security
This is the space to introduce the Services section. Briefly describe the types of services offered and highlight any special benefits or features. Encourage site visitors to learn more by exploring the full list of services offered.
Exploit Prevention
This is the space to introduce the Services section. Briefly describe the types of services offered and highlight any special benefits or features. Encourage site visitors to learn more by exploring the full list of services offered.
Theft & Code Protection
This is the space to introduce the Services section. Briefly describe the types of services offered and highlight any special benefits or features. Encourage site visitors to learn more by exploring the full list of services offered.
Why Partner
with Codyx
Keep your data safe while your team works remotely
This is a space to share more about the business: who's behind it, what it does and what this site has to offer. It’s an opportunity to tell the story behind the business or describe a special service or product it offers. You can use this section to share the company history or highlight a particular feature that sets it apart from competitors.
Let the writing speak for itself. Keep a consistent tone and voice throughout the website to stay true to the brand image and give visitors a taste of the company’s values and personality.
Reclaim more than 70% of time & resources spent on patching
This is a space to share more about the business: who's behind it, what it does and what this site has to offer. It’s an opportunity to tell the story behind the business or describe a special service or product it offers. You can use this section to share the company history or highlight a particular feature that sets it apart from competitors.
Let the writing speak for itself. Keep a consistent tone and voice throughout the website to stay true to the brand image and give visitors a taste of the company’s values and personality.
Manage your pipeline continuously with full governance
This is a space to share more about the business: who's behind it, what it does and what this site has to offer. It’s an opportunity to tell the story behind the business or describe a special service or product it offers. You can use this section to share the company history or highlight a particular feature that sets it apart from competitors.
Let the writing speak for itself. Keep a consistent tone and voice throughout the website to stay true to the brand image and give visitors a taste of the company’s values and personality.
Get started with Codyx Today
What's on offer
Stooges Web
The consumer roundtable at stooges.ai. Configure your fabric. Ask anything. Your sessions stay in your cloud storage — we never hold your data.
Stooges Junior
Same fabric architecture. Different safety guarantee. The Director reads every prompt before any model sees it. The child has no addressable channel to the safety layer — jailbreaks aren't difficult, they're structurally impossible. Backed by Jailbird AI Containment and Scruple Auditing.
Why Stooges
Stooges WThree minds, the right three, every timeeb
The fabric varies on three independent axes — model (Claude vs GPT vs Gemini vs Grok), role (Engineer vs Critic vs Drafter), personality (Sherlock vs Mad Hatter vs Cleopatra). No other multi-agent system varies on all three.
Cast morphs to fit the task
Today the Librarian leads. Tomorrow Loki reads the contract for loopholes. Next week the Coder ships your refactor. The fabric picks the right minds for the moment — and you can pick them too.
Live web search
built in
When your question needs current information, the fabric grounds itself in live sources before any model answers. No model bluffs about what happened last month.
Your data, your cloud
Stooges never holds your file content. Connect Google Drive, OneDrive, or GitHub once. Every session writes to your storage, not ours. Privacy by architecture, not by policy.
Character stooges with deep model training profiles — Sherlock, Rumi, the Mad Hatter, Cleopatra, Mark Twain, the Mad Scientist, the Skeptic — behave structurally differently than the same base models prompted with persona instructions. They function as a pseudo-Mixture-of-Experts: distinct cognitive lenses, each tuned for different aspects of a question, producing reasoning diversity that runtime persona prompting cannot match.
Persona Prompt
("Act like Sherlock")
Same base model
Character Stooge
(Deep-profile Sherlock)
Distinct cognitive lens
Why this matters
Persona prompts are commodities. "You are Sherlock Holmes, an expert detective who…" prepended to any LLM works fine, until reasoning pressure thins the persona and the base model's training mean reasserts itself.
The Stooges Hypothesis is that deep-profile character stooges are not personas. Each character is trained at depth — an authored profile, retrieval-augmented character knowledge, role-specific reasoning patterns, and tone calibration composed together — so the cognitive signature doesn't revert under pressure the way a runtime instruction does.
If the Hypothesis holds, our cast isn't theming. It's architecture. "Use the fabric" isn't a metaphor — it's a pseudo-MoE at the orchestration layer, with each stooge contributing a distinct lens that persona-prompted bases cannot replicate.
The Stooges Hypothesis
The first of two testable theses we're running through the Beta.
This one is about the cast.
Why this matters
The Hypothesis is empirically testable. We measure character-stooge output versus persona-prompted-base-model output on matched questions, same model family. We measure reasoning diversity across character stooges working the same problem. If the Hypothesis holds at Beta scale, we publish the data — including the cases where character stooges produce no measurable distinction.
VS
A Safety-Tethered Inferential Fabric System — one mid-tier featured model paired with several cheap watchers from different vendor families, structural safety upstream of inference, and topic-aware context curation — produces equal or better output than a single frontier model AND surfaces in one pass what a single-model workflow takes four-to-five iterations to find, at lower total cost.
Every query is screened before inference; the safety layer sits architecturally upstream of the model the user can address.
Safety
Policy bindings tie identity, role, classification, and context to specific inference paths.
Tethered
Featured + watchers + cross-feedback is the deliberation pattern; one structured deliberation, not many ad-hoc calls
Inferential
Claude × GPT × Gemini × Grok, current × prior-generation, frontier × fast × cheap; different training distributions catch different errors.
Fabric
Productized, audited, observable; validation receipts (~820 red-team attempts at 99%+, zero attribution errors across 1,141 turns)
System
The cost-arbitrage prediction
The Hypothesis is empirically testable. We measure character-stooge output versus persona-prompted-base-model output on matched questions, same model family. We measure reasoning diversity across character stooges working the same problem. If the Hypothesis holds at Beta scale, we publish the data — including the cases where character stooges produce no measurable distinction.
When you work through a non-trivial question with a single AI, you don't ask once — you iterate. Turn 1: headline answer. Turn 2: "but what about X?" Turn 3: Y. Turn 4: edge case. Turn 5: reconcile. Five turns is normal for analysis that matters.
The Hypothesis is that a Stooges deliberation surfaces X, Y, the edge case, and the reconciliation in parallel, in one pass — because different stooges naturally prioritize different dimensions of the question.
The math is real. The Hypothesis is that quality holds at parity or better when you run that cheaper deliberation. Adjacent academic research (Mixture-of-Models, Jan 2026) supports the direction — heterogeneous small-model ensembles matching 100B+ frontier on reasoning benchmarks, with peer-mediated correction reducing sycophancy below any individual agent. That doesn't prove Stooges specifically. Real workflow testing does. That's what the Beta is for.
Help us prove them
The Beta tests both hypotheses. Use Stooges on your real workflow. Tell us where it delivers and where it falls short. Those reports are the evidence. We publish what we find — including the workflows where the Hypotheses don't hold.
At current public API pricing, the math is verifiable:
A single frontier-model query costs roughly $0.018
A mid-tier-featured + 3-cheap-watcher deliberation costs roughly $0.012 — about 35% less
More aggressive configurations (fast-tier featured + ultra-cheap watchers) run 80-90% below single-frontier cost
The iteration-collapse prediction
If the STIFS Hypothesis holds, total resolution cost — including your time — drops roughly 30-50×. That's the claim we're testing. The cost math is fact; the parity-quality + iteration-collapse claims are what the Beta is built to validate.
The STIFS Hypothesis
The second testable thesis.
This one is about the fabric.
When you work through a non-trivial question with a single AI, you don't ask once — you iterate. Turn 1: headline answer. Turn 2: "but what about X?" Turn 3: Y. Turn 4: edge case. Turn 5: reconcile. Five turns is normal for analysis that matters.
The Hypothesis is that a Stooges deliberation surfaces X, Y, the edge case, and the reconciliation in parallel, in one pass — because different stooges naturally prioritize different dimensions of the question.
Backed by architecture, not just promises
~820
at 99%+ pass rate on the Junior safety layer — the architecture works
Red-team attempts
1,141
20-session validation battery, zero attribution errors across
audited rows
0
Every deliberation is auditable end-to-end.
attribution errors
Patent filings underway
for the orchestration, safety, and audit primitives
Stooges holds no user file content
your data stays in your cloud, by architecture
Built by Docent Technologies
Stooges sits inside a coordinated invention portfolio — TME (the hardware substrate for air-gapped frontier-model operation), Scruple (cryptographic provenance witnessing for AI-generated content), and Stooges (the orchestration layer). Patent filings underway with the Jan and Mar PCT deadlines as targets.
Stooges
Orchestration
layer
Stooges Jr
Kids-safe
fabric
Scruple
Provenance witnessing
STIFS
Enterprise Layer
(coming soon)
TME
Hardware
Substrate
