Name the fields you want. Get typed, cited, confidence-scored JSON out of any mess.
AgentFields checks every extracted field against your schema, cites it to the page it came from and sends anything below your confidence threshold to a person for review. For teams processing unstructured documents at volume.
No account. No card. No sales call.
- Every field typed, validated and cited
- A confidence score on each extraction
- Low-confidence fields routed to review
Paste anything and see the structured output
Paste an email, a ticket or a page of notes. See it turned into typed fields, with the token cost of the call. Runs in your browser. No account, no card, no call.
Typed output
{
"summary": "We are the team described as Operations and data teams in insurance, logistics, healthcare billing, legal services and… and honestly the current process is slow and confusing.",
"fields": {
"primary_topic": "billing",
"contact_email": "operations@northwind-logistics.com",
"amount_mentioned": "$14,280.00",
"sentiment": "positive",
"word_count": 113
},
"entities": [
{
"text": "operations@northwind-logistics.com",
"type": "email",
"confidence": 0.97
},
{
"text": "https://agentfields.com/docs",
"type": "url",
"confidence": 0.97
},
{
"text": "(415) 555-0142",
"type": "phone",
"confidence": 0.97
},
{
"text": "$14,280.00",
"type": "money",
"confidence": 0.97
},
{
"text": "$12,950.00",
"type": "money",
"confidence": 0.97
},
{
"text": "9.3%",
"type": "percent",
"confidence": 0.97
},
{
"text": "2026-03-14",
"type": "date",
"confidence": 0.97
},
{
"text": "Northwind Logistics Inc",
"type": "org",
"confidence": 0.72
}
],
"key_phrases": [
"billing",
"call",
"logistics",
"team",
"1295000",
"1428000",
"2026-03-14",
"555-0142"
]
}Entities (8)
- emailoperations@northwind-logistics.com
- urlhttps://agentfields.com/docs
- phone(415) 555-0142
- money$14,280.00
- money$12,950.00
- percent9.3%
- date2026-03-14
- orgNorthwind Logistics Inc
- Some entities are low confidence and would be routed to a human in production.
Take this with you
No email required. It is your result.
AgentFields — extraction · 8 entities, 8 key phrases · Sentiment: positive · 180 tokens, $0.00054 Run it yourself: https://agentfields.com
This demo runs a deterministic extractor locally, so it works with no key and no network call. It is the honest floor of the product, not the ceiling: the same typed interface is backed by the full model pipeline once you connect a key.
The problem
Extraction is where AI projects quietly die.
A model returns plausible JSON, someone spot-checks ten records, it goes to production, and six weeks later a wrong field has propagated into a billing system.
Teams have no schema enforcement, no confidence signal, no citation back to the source, and no way to tell which of the ten thousand extractions to review.
Who it is for
Operations and data teams
Teams in insurance, logistics, healthcare billing, legal services and finance who process high volumes of unstructured documents.
Developers
Developers who need reliable extraction inside their own product, with a signal they can act on.
Knowing which extractions to trust
Extraction with a contract: a schema you define, and a clear signal on every field about whether it can be trusted.
A schema with rules
Define a schema with types and validation rules, and every extracted field is checked against it.
Cited to the source
Each field is cited to the exact page and bounding box it came from.
A confidence score per field
Every field is scored for confidence, so you know which ones need a person to look.
A human review queue
Anything below your confidence threshold goes to a review queue, with the source shown side by side.
Corrections that teach
Every correction becomes training signal for that schema.
Built for high volume
Designed for teams processing large volumes of unstructured documents, and for developers building extraction into a product.
How it works
AgentFields, from the first step to the result.
- 1
Define your fields
Set up a schema with types and validation rules.
- 2
Extract
Every field comes back typed, validated, scored and cited to the page and bounding box.
- 3
Review the uncertain ones
Low-confidence fields go to a review queue with the source alongside.
- 4
Improve with every correction
Corrections feed back into the schema.
Pricing
Free for small volumes, priced per document after that, with private deployment for teams that need it.
Free
Enough to test it on your own documents.
- Typed, validated fields
- Citations and confidence scores
Per documentRecommended
Volume pricing beyond 100k documents.
- Human review queue
- Corrections as training signal
Private
For teams with deployment and review requirements.
- Private deployment
- Custom schemas with SLA
- Human-in-the-loop review seats
Prices in USD. Volume pricing beyond 100k documents.
Questions people actually ask
What comes back for each field?
A typed, validated value with a confidence score and a citation to the exact page and bounding box it came from.
What happens to uncertain fields?
Anything below the confidence threshold routes to a human review queue, with the source shown side by side.
Does it get better over time?
Every correction becomes training signal for that schema.
Is there a free tier?
Yes. 50 documents a month are free.
What does it cost after that?
0.04 dollars per document up to 100k documents, then volume pricing.
Can we deploy it privately?
Yes. 1,200 dollars a month covers private deployment, custom schemas with an SLA and human-in-the-loop review seats.
Who is it for?
Operations and data teams in insurance, logistics, healthcare billing, legal services and finance, and developers who need reliable extraction in their own product.
Why should I trust the structured-extraction demo?
You see the exact structured output and what the call cost, so nothing sits between your input and the result.
Paste anything and see the structured output
Paste an email, a ticket or a page of notes. See it turned into typed fields, with the token cost of the call. Runs in your browser. No account, no card, no call.
Open the free toolIt runs in your browser. AgentFields never sees your inputs.