TenderBrief — Procurement Notice to JSON

TenderBrief converts procurement notices and tender text into consistent, evidence-linked JSON for AI agents, procurement automation, data pipelines, and downstream analysis.

Give TenderBrief a supported procurement notice and it extracts useful commercial fields while preserving uncertainty instead of inventing missing information.

What TenderBrief extracts

TenderBrief can identify and normalize:

• Tender or notice title
• Buyer / contracting authority
• Tender reference
• Submission deadline
• Publication date
• Location
• Stated contract value
• Currency
• Value type, such as estimated value
• CPV codes
• Notice status when supported
• Requirement excerpts
• Evidence supporting extracted fields
• Missing-field warnings
• Extraction warnings and limitations

The output is designed to be easy for another AI agent, procurement workflow, ETL pipeline, database, or application to consume.

Why use TenderBrief?

Procurement notices are often inconsistent.

The same information can appear under different labels, date formats, currencies, languages, or document structures. Automated systems therefore spend unnecessary time repeatedly interpreting similar information.

TenderBrief provides a lightweight normalization layer between raw procurement information and the software that needs to use it.

Instead of returning only extracted values, TenderBrief also preserves supporting evidence and uncertainty.

For supported text and HTML inputs, evidence can include the source line and supporting quote.

For supported OCDS input, the source structure is preserved where possible.

Missing information is returned as unknown rather than guessed.

Supported input

Single notice

For the easiest use case, paste one procurement notice into noticeText.

Example:

Title: Example Network Services Contract
Buyer: Example Borough Council
Reference: EX-2026-001
Submission deadline: 15 October 2026 at 14:00 UTC
Estimated value: GBP 250000
Location: London
CPV: 72400000
The supplier must provide 24/7 network monitoring.

Batch processing

Advanced users can submit up to 50 notices in the notices array.

A batch item can contain supported procurement material such as:

• text — plain notice text
• html — supplied HTML
• ocds — a supported OCDS release object

Optional metadata can include:

• id
• sourceUrl

sourceUrl is retained as metadata only.

TenderBrief does not fetch the URL.

Languages

TenderBrief recognizes a defined set of common procurement labels in:

• English
• French
• Spanish

Supported labels include common equivalents for fields such as buyer, contracting authority, deadline, value, reference, location, publication date, and CPV code.

Language support is intentionally narrow and should not be interpreted as full natural-language understanding of every tender format.

Example output

A successful result can look like:

{
  "schemaVersion": "1.0",
  "inputId": "notice-1",
  "inputType": "text",
  "extractionMethod": "deterministic_labels_v1",
  "title": "Example Network Services Contract",
  "buyer": "Example Borough Council",
  "reference": "EX-2026-001",
  "location": "London",
  "deadline": {
    "raw": "15 October 2026 at 14:00 UTC",
    "date": "2026-10-15",
    "time": "14:00:00",
    "timezone": "UTC",
    "utc": "2026-10-15T14:00:00Z"
  },
  "values": [
    {
      "raw": "GBP 250000",
      "amount": "250000",
      "currency": "GBP",
      "valueType": "estimated"
    }
  ],
  "cpvCodes": [
    "72400000"
  ],
  "requirementExcerpts": [
    {
      "quote": "The supplier must provide 24/7 network monitoring."
    }
  ],
  "warnings": [
    "cpv_codes_not_validated_against_codelist"
  ],
  "missingFields": [],
  "status": "partial"
}

The full result can also contain:

• Field-level evidence
• Source excerpts
• Content hashes
• Processing metadata
• Missing-field information
• Explicit limitations
• Duplicate-processing identifiers

Designed for AI agents

TenderBrief is useful when an AI agent needs procurement information in a predictable structure instead of repeatedly interpreting differently formatted notices.

Potential workflows include:

• Procurement opportunity monitoring
• Tender-data normalization
• Bid-management workflows
• Procurement research agents
• Automated opportunity databases
• Cross-source procurement pipelines
• Tender screening and routing
• Structured context for downstream AI systems

TenderBrief currently uses deterministic parsing for supported fields rather than relying on a generative AI model for every notice.

This makes the output easier to reproduce and reduces the risk of unsupported values being invented.

Evidence-linked extraction

TenderBrief attempts to preserve evidence for extracted fields.

For example:

{
  "buyer": [
    {
      "line": 2,
      "quote": "Buyer: Example Borough Council",
      "basis": "cleaned_text",
      "untrustedSourceText": true
    }
  ]
}

This allows downstream systems to inspect not only the extracted value, but also the source text that supported it.

Source excerpts are treated as untrusted data, not instructions for a consuming AI agent.

Requirement excerpts

TenderBrief can identify text containing supported requirement-style language, for example:

The supplier must provide 24/7 network monitoring.

These excerpts are returned as source evidence.

They are not legal interpretations and are not determinations of bidder eligibility.

Date handling

TenderBrief normalizes supported dates.

For ambiguous numeric dates such as:

03/04/2026

you can choose:

• Auto — do not guess when genuinely ambiguous
• DMY — interpret as day/month/year
• MDY — interpret as month/day/year

The default is:

Auto

For supported date-time formats, TenderBrief can return:

• Raw source value
• Normalized date
• Normalized time
• Time zone
• UTC timestamp
• Precision
• Parsing issues

Duplicate detection

TenderBrief generates a content hash for processed notices.

Within supported workflows, this helps identify duplicate notice content and avoid unnecessary repeated processing.

Output status

Results can include statuses such as:

• structured — supported fields were extracted successfully
• partial — useful structure was extracted but warnings or unresolved validation remain
• unstructured — insufficient supported structure was identified

A partial result is not necessarily a failed extraction.

For example, a notice may be successfully parsed while still carrying a warning that a CPV code has not been independently checked against the official CPV code list.

Warnings are intentional.

They help prevent downstream agents from treating uncertain information as established fact.

Pricing

TenderBrief uses Apify pay-per-event pricing.

The primary paid event is:

notice-processed

It represents one unique procurement notice successfully delivered as a structured or partially structured result.

Invalid inputs, duplicates, and notices with insufficient supported structure are not intended to trigger the notice-processing result event.

A separate Actor-start event may still apply according to the pricing shown on the Apify Store listing.

Always refer to the Store pricing section for the current price.

Important limitations

TenderBrief is deliberately conservative.

It does not currently:

• Fetch URLs supplied in the input
• Download linked attachments
• Read PDF files directly
• Perform OCR
• Render JavaScript-heavy webpages
• Provide legal advice
• Provide procurement advice
• Determine whether a bidder is legally eligible
• Guarantee that a procurement notice is complete
• Guarantee that a procurement notice is current
• Independently authenticate the source document
• Independently validate every CPV code against the official CPV code list
• Treat missing information as proof that a requirement does not exist
• Automatically resolve every complex lot structure
• Automatically combine multiple amendments into a complete procurement history
• Reliably understand every free-form procurement notice

The quality of the result depends on the material supplied.

Free-form prose with no recognizable supported structure may be rejected rather than guessed.

Security and trust

Procurement notice content should be treated as untrusted input.

TenderBrief returns source excerpts as data only.

A downstream AI system should not interpret instructions found inside procurement text as instructions to the AI itself.

Do not include:

• Passwords
• API keys
• Private credentials
• Confidential information you are not authorized to process

Get started

For a quick test:

1. Open TenderBrief.
2. Paste a procurement notice into noticeText.
3. Leave ambiguous date handling on Auto.
4. Run the Actor.
5. Open the default dataset.
6. Review the normalized JSON and supporting evidence.

Example input:

Title: Example Network Services Contract
Buyer: Example Borough Council
Reference: EX-2026-001
Submission deadline: 15 October 2026 at 14:00 UTC
Estimated value: GBP 250000
Location: London
CPV: 72400000
The supplier must provide 24/7 network monitoring.

Automated use

TenderBrief can be incorporated into automated workflows through Apify's supported APIs, SDKs, integrations, and agent-compatible tooling.

A typical workflow is:

Procurement notice
        ↓
TenderBrief
        ↓
Structured JSON
        ↓
AI agent / database / ETL pipeline / procurement workflow

TenderBrief is designed to act as the normalization step between messy notice text and structured downstream processing.

Best use cases

TenderBrief is most suitable when:

• You already have procurement notice text
• You want standardized JSON
• You need evidence linked to extracted values
• You want missing information made explicit
• You need deterministic processing of supported fields
• You are processing multiple notices through an automated workflow

TenderBrief is not intended to replace a full procurement search platform or legal review service.

Support

If TenderBrief fails on a notice format that should reasonably be supported, open an Issue on the Actor and include:

• A non-confidential example, or
• A simplified reproduction of the problematic notice

Please do not post confidential procurement documents, credentials, personal information, or commercially sensitive material in a public Issue.