Fidelog · Checks business records against your own rules

Find the problems, see the proof, fix with confidence.

Hand Fidelog your contracts, invoices, expense claims and emails. It checks every record against the rules you set and finds problems such as overpayments, wrong tax rates or people approving their own claims. Each one points at the source line and shows the calculation. It can run on your own machines, and offline with a local model.

Get a demoLog in to the demo

These are findings from a batch of business records. Pick any one to see how it was found.

Log in to the full system

2Why it is a problem

Kestrel Ridge Storage Inc · FK-2026-00127

Paid before the contract was signed (from e-receipts)

  1. The rule, in one sentence

    Do not pay before the contract is signed.

  2. Turned into logic, worked on this record · click a number for its source

    1. Comparison13 days earlier✗ Breaks this rule

    ✓ Re-computed independently by a second program: same result

  3. Next step

    Verify the approval basis for paying first and add an explanation of the contract signing

    Only logs your decision; the finding stays as is; undo any time

3Where it is in the source

e-Receipt

  1. 1e-Receipt
  2. 2Payer: Kestrel Ridge Storage Inc
  3. 3Payee: Larkspur Catering Co
  4. 4Transaction date: 2026-01-13
  5. 5Transaction amount: $56,620.02
  6. 6Purpose: FK-2026-00127 Glacier NVMe Controller
  7. 7

Intro film

Fidelog in 3 minutes

Audit as the example: tens of thousands of documents, problems hiding between them, and how Fidelog checks every one against your rules and points each finding back to the source.

Checks in your own words

Asking a chatbot is not the same as defining a check.

Large language models are good at answering. Used to check thousands of records, they answer differently from one day to the next, nobody knows which rows they skipped, and when they are wrong they cannot point to the line. Business checking needs something else: a rule that is set once, applied to every record, with evidence for every result.

Asking a model directly

  • The same question gets a different answer next time
  • It reads the rows it happens to notice; nobody knows what it skipped
  • It sounds certain but cannot show the line or the arithmetic
  • With more material, it can only look at part of it at a time

Defining a check

  • Set the rule once; the result is the same every time
  • Thousands of rows, each one checked, with the count stated
  • Every result points to the source line, with the arithmetic recalculated on the page
  • New material is checked against the same rule automatically

For example, write

For a single purchase, total payments must not exceed the contract amount.

  1. Turned into logicAdd up the payments for each purchase order and compare the total with the contract amount; any excess is a finding
  2. After thatEach one lists the contract and payment lines with the difference worked out; you check it and log what you decide
  1. One sentence is one rule

    Describe what to check the way you would say it and give one example. Review the results first; if something is off, change the sentence and run it again. No formulas, no waiting for IT.

  2. Every result can be questioned

    Each result lists the rows, the arithmetic and the reasoning steps it used, with the source one click away. Recalculate on the page or change the rule and run it again.

  3. Grows with your business

    Rules are kept by version, with who changed what. Your team's checking experience accumulates, and a new colleague can use it from day one.

Why you can trust it

Every finding comes with proof you can inspect.

The exact source. Figures you can recompute. An independent check. You can follow how a finding was reached before deciding what to do about it.

  1. Points at the source line

    Go straight to the file, line and original wording that support a finding.

  2. Recomputed on the page

    See the source numbers and full formula, recalculated on the page for review.

  3. Checked again by a second program

    A separate program retraces the source lines and reasoning, and leaves a record you can inspect.

How one finding is made

Invoice tax rate outside the allowed bands

The rule: an invoice tax rate must be one of 13%, 9%, 6%, 5%, 3%, 1.5%, 1%, 0%.

Turned into logicTax rate = tax ÷ net amount 130.54 ÷ 1,864.91 = 7%, not one of these
SourceInvoice scan line 3 “Tax: ¥130.54” line 2 “Amount: ¥1,864.91”
Checked againRe-computed independently by a second program: same result

The rule, the calculation, the source lines and the second check form one evidence chain you can verify yourself.

Use cases

One evidence platform. Different kinds of problems.

From business records to customer conversations and compliance reviews, the same principle applies: find the issue, inspect the evidence, decide the next step.

Business records · demo available

Operations & audit

For operations, finance and audit teams checking what happened across documents and systems.

  • Inconsistent contracts, invoices and payments
  • Missing approvals or supporting records
  • Repeated supplier or process issues

Log in to the demo

Customer experience

Service & quality

For teams who need to understand recurring customer problems and review how they were handled.

  • Recurring issues in conversations and feedback
  • The original customer words behind a finding
  • Evidence to guide service follow-up and review

Start with your conversations and recordsRun one batch first

Risk & oversight

Compliance & review

For teams responsible for checking whether records and processes support the decisions being made.

  • Gaps between requirements and supporting records
  • Conflicting information across documents
  • Approval and event timelines that need review

Checks scoped to your processRun one batch first

Patterns

See the pattern behind the problem.

A single record shows one event. Related records can reveal what keeps recurring, where accounts disagree and which steps are missing.

  • Recurring issues

    The same supplier’s invoices show the same error month after month.

  • Conflicting records

    The contract price differs from the invoice; one person sits in different departments in two lists.

  • Missing steps

    An expense claim says “approved by manager”, but no approval record can be found.

  • Unusual timelines

    Paid before the contract was signed; signed off as received before the goods arrived.

What it looks like

Find the issue. Open the evidence. Take the next step.

An audit example: see how many items need you, open one to read the source, then record how it was handled.

CCorporate auditOwner

Corporate audit

45 items need attention

Next to look at

Finding
Invoice tax rate outside the allowed bandsGuilan Storage
Source
Tax: ¥130.54
Amount: ¥1,864.91
Replay
A second program reached the same result
To handle
Needs attention
Companies

Everything in the interface is clickable: tabs, findings, companies, documents.

Security

Installed on your machine. Runs offline.

Run Fidelog on your own machine or inside your network. With local models, document processing and evidence checks can work offline. External model access is optional and under your control.

  1. Keep data in your environment

    Deploy locally or on your network, with control over processing, evidence and access.

  2. Trace documents and changes

    Each document keeps a fingerprint that no longer matches once its content changes; with the activity log you can see which version was used and who changed what.

  3. Keep control of decisions

    Fidelog provides findings, evidence and next steps for your team to review and act on.

Explore deployment and data protection

Questions

The four things you'll probably ask first

How is it priced? Can we try it first?

We quote by use case, volume and deployment, listing the software, implementation and support included. You can start with a trial on a representative batch and decide after seeing the results; any trial cost is agreed up front.

Can we deploy on our own network?

Yes. Fidelog supports local and private-network deployment, with offline operation using local models. External model connections and data access follow your requirements.

How soon do we see the first result?

Pick one use case and a representative batch (say, one month of expense claims). Once we see the volume and formats, we confirm the scope and timing.

What hardware is needed?

One machine inside your network is enough to start; add more as the volume grows. Sized to your document volume.

Make Every Decision Defensible.

Bring one batch of your own material. We will show you what needs attention, the evidence behind it and how your team can take the next step.

  1. Choose one use case and a representative batch of existing documents
  2. Agree the scope, local deployment requirements and expected timing
  3. Review the findings and their source evidence together before expanding

Get a demoContact us

contact@fidelog.com