About Deniable
Deniable is a focused library of pre-generated synthetic documents for AI systems, software, and workflow testing. We make difficult inputs easier to inspect without putting production data into the test loop.
Why Deniable exists
The useful question is not whether data is synthetic or real. It is whether the input exposes the behaviour your system needs to handle, and whether your team can inspect that behaviour safely and repeatably.
The reason
Real records carry privacy, access, and operational constraints. At the other extreme, tidy fixtures can make a workflow look reliable while leaving missing context, conflicting details, and unusual structure untested.
The approach
Deniable organises synthetic documents around the behaviour a team wants to inspect. Packs are prepared in advance, documented by category and format, and made for repeatable runs rather than on-demand generation.
The human part
Every Deniable pack combines human input with creative scenario design. That involvement gives each case a reason to exist and keeps the collection focused on useful variation instead of random noise.
The boundary
Deniable documents are synthetic. They are not customer records, legal advice, financial records, or a promise that a test suite represents every production situation. The catalogue describes the scope of each release so teams can decide where it fits.
Why I built it
I built my own email parser and quickly ran into a practical problem: there was no straightforward place to buy realistic, edge-case-heavy email and document test data. The options I found were mostly enterprise-grade systems or tools that expected teams to generate and maintain everything themselves.
Deniable is the simpler route. Choose a fixed pack, use the files or the API, and spend your time testing the product instead of assembling another data pipeline. I dislike subscriptions for one-off work, so the model is built around fixed packages and prepaid API credits rather than a monthly commitment — designed as something I would buy myself.
We run a controlled local generation workflow for tighter data control and predictable production costs. Enterprise-scale cloud token usage would make this work slower and more expensive for everyone, so Deniable packages the finished, reviewed material instead. The hard part is the complete workflow around it: deciding which failure modes matter, shaping the context, reviewing the result, handling multiple formats, and packaging everything for repeatable use.
Where it fits
The same synthetic documents can support different teams. The right use depends on the workflow, the release contents, and the question you need to answer.
NLP training and evaluation for sentiment, intent, entities, clustering, and classification.
Dialogue context, interruptions, slang, and response workflows without real conversations.
Retrieval, grounding, citations, and long-context behaviour across structured documents.
EML parsing, headers, threading, encoding, and unusual message structure.
Ticket flows, routing, priorities, automations, and support dashboards.
Text aggregation, topic analysis, sentiment trends, and reporting pipelines.
Safe synthetic patterns for reviewing redaction, masking, and sensitive-field handling.
Repeatable document volumes for stress runs when the selected release supports that scale.
Clear, repeatable material for NLP, data science, and software-testing exercises.
Controlled reviews and social-style text for themes, sentiment, and trend workflows.
How we work
A case begins with a behaviour to inspect: retrieval, extraction, classification, routing, summarisation, or a workflow boundary.
Human input and creative scenario design establish the context, conflict, omissions, and format changes that make the case useful.
Each published pack is described by its category, included formats, language, package tier, and verified document count when available.
Dataset packs
Fixed dataset packs for teams that want a ready-to-download collection.
Browse datasetsAPI credits
Retrieve pre-generated cases as JSON with prepaid credits. No monthly subscription, and no generation pipeline to maintain.
BUY API CREDITS