Private professional reference
Fetch
Upload it → Ask it → Fetch it → Verify it in the original document.
Load your professional PDFs. Ask a question in normal language. Fetch finds the strongest relevant passage, tells you exactly where it came from, and opens the source page.
Ready to search
Type a standard number, clause, fire-safety measure or a plain-English defect description.
Saved on this device
Favourites
Keep commonly used documents within quick reach.
Saved on this device
Recent searches
Tap a previous search to run it again.
Saved privately on this device
Snapshots
Captioned Standards and Code page images captured from your connected PDFs. They are never uploaded by Fetch.
App information
Setup
Version 0.15.6-PROVENANCE-DEV-Q8 invariant-integrity repair development build · VFLEX PTY LIMITED
How Fetch works
Upload it → Ask it → Fetch it → Verify it in the original document.
You do not need to know the exact wording used in the document. Ask the question naturally, the way you would ask another person. Fetch searches your connected PDFs for the strongest relevant supporting evidence.
Fetch does not replace the original document. Always verify important information against the source page provided.
Install on this phone
Open this app through its web address, then use your browser’s “Add to Home screen” or “Install app” option.
Old Spec Search icon still showing?
V15 includes the correct Fetch icon, but an existing Spec Search icon will not reliably redraw itself. Keep the old icon until V15 is verified and your private text index has been exported. Then add Fetch from Safari as a second Home Screen installation, reconnect or import its private data, and delete the old Spec Search icon only after the new one is working.
Standards and documents are not included
Fetch supplies the search service only. It does not include Australian Standards, NCC/BCA publications, legislation PDFs or any other third-party documents.
- Australian Standards must be purchased or licensed separately from Standards Australia or an authorised distributor.
- NCC/BCA documents and other references must be obtained separately from their official or authorised source.
- Customers add their own lawfully obtained PDFs to Fetch on their own device.
- The customer is responsible for complying with each document’s copyright and licence conditions. Only connect a document when its licence permits that use.
Buying or subscribing to Fetch does not purchase a Standard, supply a document, replace a document licence, or grant any rights to third-party content. Always verify the current edition and exact requirement in the authorised source document.
Private PDF library
Choose PDFs from this device or Dropbox. Every newly added PDF is privately verified with a SHA-256 content digest, automatically indexed and made available to full-text search, including documents that are not yet in the built-in catalogue. They stay in this browser on this device and are not uploaded to the website.
Content-integrity protection: any older page index that is not content-hash verified is safely reset and queued for re-indexing. Existing content-hash-verified indexes remain available, so this update does not require another full rebuild.
Documents are not supplied by Fetch. Only connect copies that you purchased, licensed or otherwise obtained lawfully, and only when their licence permits this use.
Dropbox on iPhone: if a file is reported as an empty placeholder, connect to Wi-Fi or enable Dropbox cellular downloads, open the PDF once in Apple Files, or copy it to On My iPhone before selecting it.
Device storage protection
Fetch keeps connected PDFs, page indexes and snapshots in this browser. Request protected storage to reduce the chance of the browser automatically removing that private data.
Private page index
Newly added PDFs are integrity-checked and indexed automatically. Fetch now validates older connected indexes against their stored source blob, complete page coverage, fingerprints and current index metadata. Valid indexes are left untouched; stale, incomplete or retry-marked derived indexes are repaired sequentially from the existing local PDF without deleting or duplicating it. Progress is saved in digest- and generation-bound batches and resumes safely after the app reopens.
Recovery diagnostics
View non-document worker state, recovery errors and source-Blob metadata for failed or retry-marked PDFs. This read-only witness does not read PDF contents, change the recovery queue or start indexing.
Move the private search index to your phone
For faster setup, index the PDFs on the computer, export the private text index, then import that file on the phone after the exact matching PDFs are connected. The current transfer uses SHA-256 content matching, exports only complete indexes and rejects partial or filename-only imports. It contains extracted text and page numbers, not the PDF files. Keep it private and comply with each document licence.
Experimental SQLite FTS5 comparison lab
The FTS5 lab remains a separate side-by-side retrieval experiment. In r2, the normal Fetch path also includes systemic search-quality fixes for maintenance-frequency routing, specialised test-method binding and governing-clause cross-references. The lab reads the same private page index already stored on this device and builds a temporary in-memory SQLite FTS5 index for comparison.
First test on the computer. The FTS5 lab loads the SQLite WebAssembly program from jsDelivr and may use substantial RAM while indexing 30,000+ pages. It does not upload your indexed document text.
Open FTS5 comparison labExperimental Hybrid Search
Normal Fetch uses Engine A by default. This optional lab compares the same query through Engine A, B and C. Engine B uses local SQLite FTS5/BM25, and experimental Engine C combines current, lexical and local semantic candidates before the existing Fetch technical verifier and Phase 6G authority sequence.
The model and runtimes are self-hosted in this development build. Private page text and queries stay in this browser. The milestone indexes are temporary and do not alter IndexedDB.
Open Experimental Hybrid SearchCarried forward into 0.15.6-PROVENANCE-DEV-Q8 — bounded Phase 1 invariant repair only
This release includes semantic-role coverage above the proven evidence-unit engine. Search plans distinguish critical objects, application/environment context and requested properties; missing critical roles cap confidence; selection/design queries route away from routine-service material; test switches remain a discrete object rather than collapsing into generic testing; drafts and historical editions are ranked more cautiously; and late private-index enrichment cannot replace a stronger visible result on raw lexical score alone. The sibling-subclause, heading-anchor and switchboard protections remain intact.
- Clause-sized evidence binding prevents unrelated provisions on the same page from combining into a false answer
- Requirement questions beginning “what is/what are” remain requirement searches when measurement, threshold, clause or mandatory-language cues are present
- Morphology and light function-word tolerance accepts wording such as “mounting heights” and “above the floor” without weakening object binding
- NCC/BCA currency is tracked separately by Volume One, Two and Three
- Exit-door requirement searches prefer code provisions and demote offence/enforcement material unless that material was explicitly requested
- Compact private-page fingerprints use whole-page Bloom recall plus beginning/middle/end samples so late clauses remain discoverable
- SHA-256 content identity binds each PDF, page index, imported index and learned answer to the exact file bytes
- Different bytes under the same filename invalidate stale page text and derived links before automatic re-indexing
- Complete-only, digest-bound private-index export/import rejects partial and filename-only matches
- Evidence-first page ranking uses word boundaries, controlled morphology, proximity and substantive requirement wording
- Contents, indexes and OCR-spaced navigation pages are hard-excluded unless explicitly requested
- Table-focused questions prefer the actual requirement table; AS/NZS 3000 Table 3.6 outranks nearby general clauses
- Applicable-date searches validate real dates and show later editions only as clearly marked references
- Search-priority indexing moves likely matching unindexed PDFs to the front of the queue
- Separate local feedback confirms the evidence page independently from the generated site answer, so a correct page is never rejected merely because an answer summary is wrong
- Saved feedback is single-use: once confirmed, the choice buttons disappear and cannot be clicked again or accidentally reversed
- Ceiling-fan searches bind the 400 mm measurement to the fan-blade wording instead of selecting the nearby 900 mm air-supply-opening measurement
- Broad refrigeration-room searches prefer the start of Section 6.6 and Clause 6.6.3 before later specialised clauses
- Corrupted selectable text is replaced with a visual-PDF verification warning instead of being presented as a reliable quotation
- Strict document identity prevents incidental references inside another Standard from becoming the answer source
- Essential-term and source-identity checks cover underground-wiring and refrigeration-room wording variants
- Current 177-record catalogue copied from AFSS Report Builder v1.1.62F
- Field-answer search that targets the actual indexed page instead of stopping at the document title
- Site-first result workflow: Find answer first, then open the PDF, view details or copy the reference
- After an answer is found: Open matching PDF page first, Save matching page snapshot second
- Detailed-reference dialog now puts the matching PDF action before copy and labels unindexed documents correctly
- Improved natural-language ranking, intent routing and common misspelling handling
- Targeted search for practical defects such as detector spacing, hose reels, extinguisher travel distance, fire-door closing and sprinkler-valve access
- Structured searches for standard, clause, applicable year/date, source, group, exact phrases and exclusions
- Catalogue metadata and available sample text
- Document-type and individual-document filters
- Voice-search control where supported
- Safari Home Screen launch and offline-cache redirect correction
- Device-local favourites and recent searches
- Private device/Dropbox PDF selection and document connection
- Clearer document-only and clause-page status
- Red Clear on the left and green Search on the right
- Automatic private full-text indexing and search registration for every newly added PDF
- Private page-by-page PDF text indexing, clause/figure heading detection, PDF/printed page labels, answer extracts and open-at-page controls
- View detailed reference and Copy detailed reference actions on every result
- Automatic exact-clause search inside indexed connected PDFs; manual “Find exact answer now” remains only as a fallback while a PDF is not yet indexed
- Private index export/import so a computer-generated text index can be transferred to the phone
- Resumable PDF indexing with progress saved every five pages
- Matched, unmatched and index-error PDF audit filters
- Safe automatic linking by Standard number + edition for differently named connected copies; manual linking remains for ambiguous cases
- Requirement searches reject definition-heavy pages unless a definition is actually requested
- No-date searches favour the newest connected/catalogued edition and clearly label older editions as historical
- No · show next keeps the result stable while Fetch looks for the next stronger evidence page
- Browser storage capacity display and protected-storage request where supported
- Restrictive content security policy prevents the app from sending connected PDF content to external websites
- Caption-required Standards / Code page snapshots saved privately on this device
- Better Australian Standard classification and cleaner result samples
- Fetch branding and “Upload it → Ask it → Fetch it → Verify it” workflow
- Clear customer notice that Fetch includes no Standards, NCC/BCA PDFs or other full documents
- Versioned Fetch Home Screen icons for clean iPhone re-installation
- Resilient bulk PDF importing that continues after an unreadable Dropbox file and reports every failure
- Topic-aware search routing for maintenance, installation and NCC/BCA requirement questions
- Versioned app assets prevent older cached CSS and search logic from mixing with V15
- Offline app shell and catalogue cache
Private means one device
Connected PDFs, the page index and snapshots stay in this browser on this device. They do not automatically copy to your phone or another computer.