Clear answers, including the limits.

LeakHunt AI is a financial issue detection tool for small and medium-sized businesses. It analyzes invoices, receipts, supplier price lists, and expense spreadsheets to surface potential duplicates, price changes, unusual spend, and savings opportunities — with the evidence behind each finding.

Accounts, isolated organization workspaces shared across teams with roles and invitations, CSV and XLSX imports, deterministic detection, finding review, recorded confirmed and realized savings, CSV reports, the grounded analyst, and Free-plan usage limits are available now. PDF, image OCR, paid plans, and payments are not enabled yet.

Owners, finance leads, bookkeepers, and operations teams at small and medium-sized businesses who can review supplier invoices and expense records but do not have a full-time audit team.

CSV and XLSX imports with column mapping, validation, and confirmation; duplicate invoice and duplicate payment checks with payment evidence; supplier price increases; unusual expense flags; findings review; a recorded savings ledger with CSV exports; and a grounded analyst. PDF and image OCR remain planned, as do paid tiers.

LeakHunt compares supplier, invoice number, invoice date, amount, currency, and line items where available. Exact duplicates match on the strongest identifiers. Probable duplicates match on a combination of signals with some differences, such as a nearby date or slightly different line items. The finding shows which signals matched.

No. It means two similar invoice records exist. A possible duplicate payment is only flagged when separate payment records support that conclusion. The two findings are deliberately kept separate.

The same product or service is compared across invoices. SKU or product identifiers are preferred, with normalized descriptions only as a fallback. Comparisons require compatible currencies and units of measure, and show previous price, current price, absolute change, percentage change, and comparison dates.

LeakHunt uses transparent rules and basic statistics against your own historical data, such as a large deviation from a supplier's normal amount or a new recurring charge. The tool requires enough history before comparing and always shows the reason for the flag. An anomaly is not proof of fraud or an incorrect payment.

A finding is a lead based on recorded evidence. Potential savings are estimates with visible assumptions. Only an authorized user can change a finding's review status or record a confirmed amount and a realized recovery, and recorded amounts are kept separate from estimates so the same money is never counted twice. A recorded amount is what a reviewer entered; the app has no payment or bank data to verify it against.

CSV and XLSX files containing invoice, expense, supplier price, or payment records. The columns are mapped and validated before an import is confirmed. The original file is not retained after import; only validated rows, the filename, and a file fingerprint are stored.

No. Every customer record carries an organization ID, server-side membership checks run before reads and writes, and Row Level Security protects the database. Cross-organization access tests are part of the test suite.

Yes. An owner or administrator invites a teammate by email address and receives a one-time link to send yourself; the link carries a secret that is stored only as a hash, expires in 7 days, and is accepted only by the invited account. Owners change roles and transfer ownership, administrators remove members, and anyone but the owner can leave. Roles and seat counts are enforced by the database, and email delivery of the links is not connected yet.

In PostgreSQL through Supabase, with organization-level isolation and Row Level Security. Application code does not use a Supabase service-role key, and original uploaded files are not retained. Hosting region, retention, and backup details will be published before commercial launch.

Core detection does not require an AI provider. The optional grounded analyst runs locally by default and only explains stored findings. If an administrator configures a server-side external provider, the selected finding, counts, estimates, and up to 10 open findings are sent with the question. External answers are checked for unsupported amounts and claims, with fallback to local mode. Uploaded documents are treated as data, never as instructions.

Yes. An owner can delete the organization, which permanently removes its imports, rows, findings, reviews, and recorded savings, and an account can delete itself once it belongs to no organization. Both are confirmed by typing the exact name and are enforced by database functions. Legally required retention has not been assessed for a commercial launch.

The Free plan is available now with monthly limits enforced by the database. Two paid tiers are planned at provisional prices of $49 and $149 per month. Those prices are not live offers, and no payment can be made while billing is disconnected.

Paid plans will open only after the core product is validated and secure. A payment provider that supports the business jurisdiction and intended customers will be integrated with verified server-side webhooks. Paid access will never be unlocked based only on the browser's claim.

The Free plan allows 2 imports, 2,000 confirmed valid records, and 1 member per UTC calendar month. An import counts when it is uploaded; confirmed records count valid rows from imports confirmed during the period. Invalid rows are kept for audit but excluded from the record limit. Unused allowance does not roll over.