No refusal theater.
Hard questions get real answers. When a request is legitimate, Agent19 treats it like work, not a policy problem.
MODEL 01 · LIVE NOW
Do the work other models won't.
Agent19 is an unrestricted AI model for developers and teams doing serious work. Security research. Forensic analysis. Age-restricted platforms. The hard problems everyone else refuses. Governed by your rules, not a policy layer. Running today.
MODEL 01 · LIVEunrestricted core
YOUR RULES LOADEDgovernance at your edge
NO REFUSAL LOGreal work, real answers
LIVE
Model 01 is running today. Not a waitlist.
UNRESTRICTED
No policy-layer refusals on hard work.
INDEPENDENT
Built by builders, governed by you.
THE MODEL
Useful where others tap out.
Most models get squeamish exactly where the work gets serious. Agent19 was built for that moment: the security audit that needs a real answer, the forensic write-up, the age-gated platforms general models moralize about, the dataset nobody else will generate. No lectures. No dead ends. A model that does the job.
Your governance layer.
Unrestricted at the model, governed at your edge. You set the rules for each product and enforce them in your stack. Allow, block, log. Your call.
Built for production.
A working model, live now, built for teams shipping real software. Not a demo that folds under load.
USE CASES
Generate the data other models won't.
The datasets that move work forward are the ones general models refuse. Adversarial examples. Red-team transcripts. Edge-case conversations. Domain corpora. Agent19 generates them.
Security research
Tooling, write-ups, and test payloads without the intermission of refusals.
Investigative + forensic work
Analysis and drafting for people whose job is the hard stuff.
Age-restricted platforms
Age-gated, consent-first products for adults, built without a model that moralizes.
Training data
Edge-case corpora, adversarial sets, and evaluation data other providers won't touch.
GOVERNANCE
Unrestricted doesn't mean unmanaged.
The model ships unrestricted. Your product doesn't have to. Agent19 is built for teams that enforce their own policies, per product, per user, per market. Your rules run at your layer, where they belong.
YOU SET THE RULES
YOU ENFORCE THEM
YOU ANSWER TO YOUR USERS, NOT OUR POLICY TEAM
DATA
Your data stays yours.
Your prompts are your business. Defaults favor you: no ad targeting, no data brokerage, no training on your traffic. Retention is yours to configure, and the details get published plainly when access opens.
RETENTION AND PROCESSING DETAILS ARE PUBLISHED WITH PUBLIC ACCESS.
STACK
Built for the tools you build with.
Agent19 is designed to slot into the workflows you already run: agent frameworks, research pipelines, content platforms, internal tooling. If your stack talks to a model, it can talk to Agent19.
> agent19.complete({
model: "agent19-01",
input: "the work other models won't",
rules: "yours"
})
→ live. no refusal. no lecture.ACCESS
Simple when it opens.
Model 01 is live for early builders today. Access expands in stages, and pricing ships with it. Usage-based, published plainly, no seat math.
USAGE-BASED
Pay for tokens, not seats.
PUBLISHED PLAINLY
Public pricing with public access.
NO THEATRICS
No enterprise-only maze.
Frequently asked questions.
01/ What is Agent19?
An independent AI company shipping an unrestricted model for serious work. Model 01 is live today. The direction is a broader model and inference platform.
02/ Who is Agent19 for?
Developers and teams whose work gets refused elsewhere: security research, forensic and investigative work, age-restricted platforms, training data.
03/ Does unrestricted mean unsafe?
No. The model ships unrestricted at its layer. You govern at yours. Most teams ship Agent19 behind their own policy engine. That's the design.
04/ What can the model do?
General-purpose model work: chat, completion, analysis, drafting, and generation, with far fewer refusals on legitimate hard work.
05/ Can I generate training data?
Yes. Adversarial sets, edge-case corpora, evaluation data. The datasets other providers won't produce.
06/ How do I control what it allows?
Your layer, your rules. Set policy per product, per user, per market. Agent19 doesn't override you.
07/ Do you train on my prompts?
No. Defaults favor you: no training on your traffic, no ad targeting, no data brokerage.
08/ How does pricing work?
Usage-based and published plainly. Pricing ships with public access. No seat math, no enterprise-only maze.
09/ How do I get started?
Model 01 is live for early builders. Access expands in stages. This page is the source of truth.
10/ What comes after Model 01?
A broader model catalog and an inference platform. The direction serious AI infrastructure is going, built our way.
Run the model that does the work.
Unrestricted at the core. Governed by you. Live today.