Provance Synthetic
Training conversations that carry their own evidence.
Turn your domain's knowledge into gold standards and large volumes of training conversations. Every conversation is de-identified, QA-checked and safety-reviewed, and anything uncertain waits for a person, before it can be exported.
How it works
From domain to dataset in one guided flow.
Each step builds on a fixed version of the one before, so every conversation records exactly which ontology, scenarios and gold standards it came from.
- 01
Domain
Describe the work your AI agent will do and the rules it must follow. - 02
Ontology
An AI-built map of the problems, causes and relationships in your domain, which you edit. - 03
Scenarios
Situations that cover the problem space, with coverage shown as you go. - 04
Gold standards
Reference conversations that pass every check: the anchors for everything generated. - 05
Runs
Batches of conversations, each de-identified, QA-checked and safety-reviewed.Reviewed - 06
Export
Only what passed, in your format, to your destination, with its evidence.Sealed
What every conversation goes through
Checked before it counts as yours.
- Personal details removed before storage or any model call
- Quality rules for structure, content and your domain's own requirements
- An AI safety review of every assistant reply against a versioned policy, including the escalation rules your domain needs
- Uncertain items held for an owner or admin to approve or reject
- An export gate: nothing rejected, held or carrying personal details can leave
- A plain-language assurance summary per conversation and per run
You see what you got
Every generated conversation counts towards your allowance, including ones the checks reject, and the QA pass rate is shown on every run. No hidden filtering, no surprises.
The moment of value
The free preview builds your domain's ontology and three gold standards with the full QA report, plus a 10-conversation sample. You see your own domain pass every check before you pay.
Export
In the format your stack already uses.
Download, send to your S3 bucket (including through an IAM role in your own account), Google Cloud Storage, Azure or Hugging Face, with train, eval and test splits.
- OpenAI
- Anthropic / Bedrock
- Vertex / Gemini
- OpenAI Batch
- Llama
- ShareGPT
- Alpaca
- ChatML
- Hugging Face
- Parquet
- CSV
- Markdown
- Croissant metadata
Plans
Sized by domains and monthly conversations.
Never priced per document. Synthetic always runs on Provance-hosted models, chosen and evaluated for each role.
Free preview
See gold standards for your own domain before you pay.
Free
- 1 domain with an AI-built ontology
- 3 gold standards with the full QA report
- A 10-conversation sample to review
- Export unlocks with a paid plan
Starter
One domain, steady monthly volume.
$299 /month
Plus GST
- 1 active domain
- Up to 3,000 conversations a month
- Up to 100 new gold standards a month (fair use)
- 1 run at a time
- Every export format and destination
Recommended
Growth
Several domains and more volume.
$799 /month
Plus GST
- 3 active domains
- Up to 10,000 conversations a month
- Up to 300 new gold standards a month (fair use)
- 3 runs at once
- Quality-gate reports, webhooks and API
Scale
Many domains at production volume.
$1,999 /month
Plus GST
- 10 active domains
- Up to 30,000 conversations a month
- Up to 800 new gold standards a month (fair use)
- 10 runs at once
- Team roles, audit and provenance exports
Enterprise
Contracted domains and capacity.
From US$40,000 a year
- Domains and capacity set by contract
- Up to 50 runs at once
Top-ups
Extra conversations for the current period on any paid plan, bought in one click on your saved card. Unused pack conversations carry over once, then expire. Auto top-up is off unless you turn it on, with a maximum you set.
| Pack | Conversations | USD | AUD |
|---|---|---|---|
| Top-up pack | 3,500 | $250 | A$385 |
| Auto top-up step | 1,400 | $100 | A$154 |
Premium assurance add-on
Not yet availableHigh-risk conversations generated on Claude Opus 5.5. It goes on sale once that model passes its qualification for the role; until then it cannot be bought.
| With plan | USD /month | AUD /month |
|---|---|---|
| starter | $99 | A$149 |
| growth | $299 | A$459 |
| scale | $899 | A$1,379 |
All prices exclude GST; 10% GST is added at checkout. Over your domain limit after a downgrade? Nothing is deleted: new runs pause until you archive a domain or change plan.
FAQ
Synthetic questions.
Can I use my own model key for Synthetic?
No. Synthetic always runs on Provance-hosted models, because Provance chooses and evaluates the model for each role (generation, gold standards, safety review), and adopts a new model only after it passes the same evaluation.
What counts towards my monthly conversations?
Every conversation generated counts, including ones the QA checks reject. Your results show the QA pass rate, so you can see exactly what you received. Rejected conversations can never be exported.
What happens when I reach my monthly limit?
A run already in progress finishes. New runs wait until you add a top-up pack, upgrade, or your allowance renews on your billing date. You can also turn on auto top-up with a maximum you set. There are no surprise overage charges.
See a gold standard for your own domain.
The free preview builds your ontology, three gold standards with the full QA report and a 10-conversation sample. Export unlocks with a paid plan.