Documents
Chat with documents and spreadsheets in plain English.
Document Intelligence
“Your team works across PDFs, Word, slides, Excel, CSVs, scanned invoices, and websites — and every format needs a different tool, a different person, or a VLOOKUP that breaks at midnight. Important answers live in the files you already have. You just need to ask them.”
CostTrail reads every business document your team touches. Upload PDFs, Word, PowerPoint, Excel, CSV, images, or paste a website URL — then ask anything in plain English. Behind the scenes, prose gets indexed for semantic search, spreadsheet rows get loaded into a real SQL engine, and invoice fields get structured automatically — so "sum revenue by region" returns an exact answer like a database, "what does our termination clause say?" returns the cited paragraph, and "show all Salesforce invoices over $5,000" returns clickable invoice cards. Every answer carries its source.
Who uses it?
Real people with real problems this solves.
Finance / Ops Lead
Before: Excel formulas break, pivot tables get stale, "how much did we spend on Sales Cloud last quarter?" needs a manual VLOOKUP across a dozen invoices
With CostTrail: Ask in plain English — real SQL runs on your spreadsheets and line items are read from every invoice
Legal Team
Before: Finding a specific clause across hundreds of contracts takes hours of manual searching
With CostTrail: Ask "what are the termination clauses across all vendor contracts?" and get cited answers in seconds
HR Manager
Before: New hires ask the same policy questions repeatedly — each answer takes 20 minutes
With CostTrail: Upload the handbook once — everyone queries it self-serve with page-level citations
Compliance Officer
Before: Audit prep means reading regulatory documents end-to-end to find the right sections
With CostTrail: Query across all compliance documents simultaneously; answers cite the source page
Operations Lead
Before: SOPs, downtime logs, and shift reports live in separate spreadsheets, PDFs, and folders
With CostTrail: One place to ask — "median downtime per shift in week 14" works on the Excel, "what does the SOP say about lockout?" works on the PDF
How to connect
Upload any file (PDF, Word, PowerPoint, Excel, CSV, text, Markdown, images) or paste a website URL
- 1Upload your files — PDF, Word, PowerPoint, Excel/CSV, images, or paste a URL. CostTrail reads each one and figures out its structure.
- 2For spreadsheets and CSVs, it builds a real column schema (types, samples, distributions) so questions can be answered with SQL, not guesses.
- 3For PDFs and Office files, it extracts text + tables + form fields (invoice number, total, due date) so both prose and field-level questions work.
- 4Ask in plain English — every answer is cited (page number, row count, matching documents, or invoice number with a one-click Open button).
- 5Open Visual Dashboards on any tabular file for auto-suggested charts, or build your own from a prompt or column picks.
What it replaces
Every feature exists because something was broken before.
Every business file, one tool
Tabs across Excel for spreadsheets, Adobe for PDFs, an OCR tool for invoices, and a search bar that doesn't work
Accepts PDF, Word (.docx/.doc), PowerPoint (.pptx/.ppt), Excel (.xlsx/.xls), CSV, text, Markdown, images (PNG/JPG/WEBP/GIF), and website URLs — all queryable from one chat box
Plain-English Excel formulas
VLOOKUP, SUMIF, COUNTIF, AVERAGEIF, INDEX/MATCH, and pivot tables break when columns move, formulas point to the wrong cell, or someone deletes a row
Ask "sum revenue by region where status is paid" or "average downtime per machine" — runs as real SQL against your spreadsheet rows. No #REF!, no broken references, no rebuild when the data updates
SQL engine for spreadsheets
LLMs hallucinate numbers when forced to total a 50k-row CSV — and bury the wrong answer in confident prose
XLSX and CSV rows are loaded into PostgreSQL with type-classified columns. Aggregations (sum, avg, median, percentiles, stddev) run as SQL — exact numbers, every time
Visual Dashboards
Building charts in Excel takes 20 minutes per chart and nobody shares them
Open any tabular file to see auto-generated charts based on its columns (time series, top-N breakdowns, KPI tiles). Build custom charts from a prompt ("revenue by region") or pick columns directly. Bar, line, pie, KPI, scatter, table — all in the browser
Invoice & form intelligence
Field-level questions on invoices, contracts, and forms force someone to open every PDF
Vision-based extraction reads invoice number, date, total, vendor, line items, due date, tax ID, and more — automatically. "Show all Salesforce invoices over $5,000" returns invoice cards with Open buttons; "how much on Sales Cloud across all invoices" reads line items and totals the matches
Cross-document queries
Answers often span multiple files — manually opening 30 PDFs to compare clauses is a half-day of work
Query across 100+ documents in a single question. The system retrieves only the relevant chunks per file and assembles the answer with per-document citations
Citation-backed answers
AI answers without proof can't be used for legal, compliance, audit, or finance decisions
Every answer includes its source: page number for documents, row count + filters for spreadsheets, invoice number + Open-PDF button for financials. One click to verify
Data quality safety net
A spreadsheet with mixed currency symbols, blank rows, or 30% nulls quietly poisons every aggregate you compute
Ingest detects and reports duplicate rows, empty rows, repeated headers, mixed currency symbols, high null rates, and outliers per column. You see what was cleaned before you trust the answer
Website ingestion
Knowledge bases, documentation sites, and policy portals live on the web — you can't paste a URL into Excel
Paste a URL — CostTrail crawls up to 10 same-domain pages and indexes the content. Documentation becomes queryable without manual export
Saved questions + suggestions
You ask the same 5 questions every Monday and they live in your scratch notes
Star any answer to save it; on first upload, CostTrail suggests 10–30 starter questions based on what the document actually contains
Secure processing
Sensitive contracts and legal documents can't be uploaded to public AI tools
Private Mode keeps PDFs and text on your machine — only relevant snippets and your question reach the AI for the answer
Questions you can ask right now
Plain English, answered from your own data.
Industries using it
Same tool, different industries.
Legal
Hundreds of contracts per client, each 50+ pages. Finding specific clauses during due diligence or litigation takes hours of manual search.
“What are the termination clauses across all our vendor contracts?”
Healthcare
Clinical protocols, insurance guidelines, compliance documents scattered across folders. Nurses and doctors can't find answers to procedure questions quickly.
“What is the protocol for administering medication X to patients with condition Y?”
Education
Curriculum documents, accreditation handbooks, policy manuals hundreds of pages thick. Teachers re-read them repeatedly to find specific rules.
“What is the procedure for handling student appeals in our grading policy?”
Manufacturing
Technical manuals, equipment documentation, safety SOPs thousands of pages thick. Operators waste time searching PDFs when something goes wrong.
“What are the calibration steps for the CNC machine model X?”
Financial services
Regulatory filings, compliance docs, policy manuals constantly updated. Compliance officers manually verify every change between versions.
“What changed between the 2024 and 2025 AML compliance manual?”
Real estate
Lease agreements, property disclosures, HOA documents. Quick answers about tenant rights or property rules require re-reading full documents.
“What are the pet policies across all rental properties we manage?”