Preparing a tender is a single, per-tender price shown at /pricing (GET /v1/pricing). Finding and matching tenders, tender alerts, every document checker, API keys, and the MCP connector are free.
1 September 2026
Plagiarism checker: find matching published text, with sources
New tool: /plagiarism-checker (API: POST /v1/check-source-overlap; MCP: check_source_overlap on /mcp and the dedicated /mcp-plagiarism server). Paste text, a public URL, or upload a PDF/.docx — it searches the public web and reports every exact match as a quote confirmed against the fetched source page, with the URL and the position in your text.
Two evidence tiers, never mixed: exact matches — deterministic, verifiable by opening the source — ship today. The second tier, clearly-labelled model judgements on possible paraphrases, is disabled until its measured evaluation passes its precision gate; results report exact matches only and say so honestly.
Honest by construction: it reports overlap, never accusations — no match is never a certificate of originality, and the public web is not an academic database. 3 credits per completed check; an abstention or a search outage costs nothing, and the outcome field in every response is the receipt.
31 August 2026
Free tier becomes a 7-day trial; new paid credits get a 90-day life
The recurring weekly allowance is replaced by one free trial per account: 200 credits for 7 days, no card, starting on your first metered action. When it ends, metered tools pause until you buy a pack — from $5, one-off, no subscription.
Credits purchased from today are valid for 90 days from payment. Credits purchased before today keep not expiring. Spending order protects you: trial first, then the purchase closest to expiry, then any non-expiring balance.
Metered use now requires a signed-in account, and new API keys belong to your account and spend its balance — one identity for keys, credits and history. Browsing tenders, result reads and the sandbox stay free with no account.
The API tells you where you stand: /v1/pricing publishes the trial and expiry terms, /v1/usage reports trial state and your nearest expiry, and a 402 carries the cost, your balance and the checkout link as structured fields.
22 August 2026
Free tier: one allowance, 50 credits a week
Signing in (or using a free API key) now gives 50 credits a week, the same size as the anonymous allowance. What changes is whose it is: an anonymous allowance is metered per IP and shared with everyone behind that address, while a signed-in one is yours alone — and it carries your history and holds any credits you buy.
Nothing else moves: per-page extraction pricing, the weekly renewal, and the anonymous API and MCP tiers are unchanged.
21 August 2026
classify and redact fold into extract
The standalone /classify and /redact endpoints (and the classify_document and redact_pii MCP tools) are retired, as announced. Their capabilities live on /extract: options.classify returns the full classification verdict (nested under classification) and options.redact returns the masked PII inventory plus redacted text.
Price parity, kept honest: a capability-only call — options.classify and/or options.redact with no fields, template, or other option — skips field extraction entirely and costs exactly what the retired tools charged: classification 1 credit, redaction 1 credit per page.
The /classify and /redact pages on this site keep working — they now call /extract underneath.
Explicit options on /extract, every one off by default: whole tables as rows, a bounding box for every value the PDF’s own text layer can locate (grounding), document layout (headings, fonts, columns, reading order), link annotations, embedded-image positions, retrieval-ready chunks, and page-splitting for files that hold several documents. Asking for an option a server does not support is an error naming it — never a silent no-op.
Deterministic first: on a born-digital PDF, tables, layout, links, figures and chunks are read from the document’s own coordinates — exact cells, merged-cell spans, no model call and no charge for those passes. A scan falls back to the vision model where possible, and every result says which mechanism produced it. The new free endpoint GET /v1/extract/capabilities is the machine-readable contract for all of it.
Chunking ships six strategies (by section, page, character budget, recursive, per element, and hierarchical parent/child for small-to-big retrieval). Every chunk knows its heading path, page range and position, and by default a table is never sliced down the middle.
Pricing is now per page: 1 credit per page a model actually reads (minimum 1), resolved before the call runs so the cost is predictable. Deterministic work is free. The free tier renews weekly — 50 credits anonymous, 150 with a key.
The classifier now recognises more types (contract, identity document, utility bill, council rates notice, insurance certificate, superannuation statement, medical certificate) and returns a label — its own open-set name for the document — even when the closed-set verdict abstains.
Deprecation notice: /classify and /redact are superseded by /extract with options.classify and options.redact — the same information, plus fields, in one call. Both endpoints keep working unchanged and will be removed only after usage has moved.
17 June 2026
Adverse media & sanctions screening
New tool: /adverse-media — screen a person or organisation for negative news and sanctions/PEP exposure. A deterministic layer matches against consolidated global sanctions & PEP lists; a second layer scans public news and classifies adverse coverage (fraud, money laundering, bribery, sanctions, enforcement).
Built against the #1 screening failure — same-name people: sanctions matches are corroboration-gated (a name-only hit is flagged “possible”, never confirmed), and same-name news articles about other people are disambiguated out and listed separately. Add a date of birth, country, or aliases to sharpen the match.
POST /v1/adverse-media over REST; screen_adverse_media as an MCP tool on the full-suite server. Screen by name (with optional disambiguators) or pass an identity document to read the subject from it. Screening coverage, not a determination — a hit means “review”, and “nothing found” is not a clean record.
17 June 2026
PII detection & redaction
New tool: /redact — find and mask personal data in a document. A deterministic pass catches structured identifiers (emails, payment cards, US SSNs, India PANs, AU ABNs); a vision pass catches names, addresses, and dates of birth.
Privacy by design: detected values are masked in the response (the raw PII is never returned), and a redacted copy of the text is included when the document has a text layer. The document is never stored.
POST /v1/redact over REST; redact_pii as an MCP tool on the full-suite server. Detection coverage, not a guarantee — review before relying on it for compliance.
17 June 2026
Document-pack completeness — does the pack have what it needs?
New tool: /check-pack — give it a set of documents and a checklist, and it tells you which required documents are present and which are missing. Pick a named scheme (income proof, lending pre-qualification, rental application) or define your own requirements.
Built to stay cheap: it only identifies each document’s type (one page-1 read, hash-cached) — never full extraction — so a 50-page bank statement costs the same as a one-page one, capped at 12 documents per pack.
POST /v1/check-pack over REST; check_pack as an MCP tool on the full-suite server. Completeness coverage, not approval — that the right document types are present, not that any is genuine or that an application is approved.
17 June 2026
Field extraction — pull structured values from any document
New tool: /extract — a vision model reads a document and returns the fields you ask for. Say what to pull as an ad-hoc field list, a template (payslip, tax invoice, bank statement, receipt), or let it auto-detect the document type and pull the right fields; an unrecognised type falls back to every labelled field.
Reads checkboxes, and finds a named value even when it sits inside a merged-cell or nested table, across up to a few pages per document. It does not return tables as rows. A field it can’t clearly see comes back as not_found — it abstains rather than guessing.
POST /v1/extract over REST; extract_fields as an MCP tool on the full-suite server. Extraction, not verification — values are what the document shows, never proof it’s genuine.
16 June 2026
Identity check — the Australian 100-point check
New tool: /identity — verify a person’s identity from a set of documents. A reader identifies each document and the fields it shows; a deterministic engine tallies them against the AFP 100-point check or the AUSTRAC safe-harbour rules and tells you whether identity is established (and exactly what’s missing if not).
POST /v1/identity-check over REST; verify_identity as an MCP tool on the full-suite server.
Identity coverage, not a forgery judgment — abstains on documents it can’t read confidently, and never counts an income document (e.g. a payslip) toward identity.
12 June 2026
Financial document classifier + one MCP server for everything
New tool: /classify — financial document-type classification (payslip, tax invoice, bank statement, salary certificate, PAYG summary, receipt) with country and confidence. Abstains rather than guesses.
POST /v1/classify over REST; classify_document as an MCP tool.
One MCP server now carries the full suite: add /mcp once and get every tool. Scoped servers (/mcp-aitext, /mcp-verify) keep working for existing installs.
12 June 2026
API keys + usage metering
Free self-serve API keys (instant, no account): your own 50 documents/day quota instead of sharing the per-IP one.
Bearer authentication on every endpoint, REST and MCP alike.
GET /v1/usage — metered usage read from the same counters that enforce the quota.
12 June 2026
Developer documentation
New docs: quickstart, authentication, full API reference, MCP integration guide, errors & limits.
Use-case guides and this changelog.
11 June 2026
Readable fact-check links
Fact-check permalinks now use the document’s own title as the URL slug (hash-named files no longer produce hash-looking links).
10 June 2026
URL intake everywhere + agent prompts
Every MCP tool now accepts a public url — agents verify a document with one short string instead of encoding it.
Fact-check results from MCP calls return a shareable permalink.
MCP servers publish ready-made prompts (e.g. "Fact-check a report URL") that appear natively in supporting clients.
AI-text detection accepts {"url": ...} over REST.
Security hardening across the URL fetcher and upload paths.
8 June 2026
Public fact-check permalinks
Every fact-check persists under a content-hash id with a public, shareable /fc/<slug>-<hex> page.
Fact-check accepts a report URL directly — paste a link instead of uploading.
Streaming (SSE) keeps deep checks alive through proxies.
5 June 2026
Scorecard results
Redesigned results on all four surfaces: web tools, PDF report, Markdown report — score dial, verdict banner, per-component bars.
3 June 2026
Three MCP servers in the official registry
Document forensics (/mcp), AI-text detection (/mcp-aitext), and reference fact-checking (/mcp-verify) are live in the official MCP registry — hosted, no install.
New tools and pages: AI detector and Fact check, with feedback on every result.
May 2026
Stipple launches
Document verification warrants: forensic inspection with risk bands, shareable records, and audit-ready PDF reports.