Decision tracing for Node.js and PythonShow your work.

Every price, approval and AI answer your software gives - traced step by step, readable by your support team, your customers and your auditors.

Get started free

Every decision on the record. Its inputs and outcome, tagged with your own IDs so you can find it later.

A credit assessment traced in the Recount dashboard: its fields, every step with its timing, and a request with its authorization header redacted.

A lender’s branded Recount portal summarising its failed and completed runs over the last 7 days.

Credit decisions

8,412 records, kept for 25 months

Find an application IDLast 90 daysFilter

Export CSV

Application ID	Lender	Product	Amount	Credit score	Decision	Reason
APP-554559	Tidewell Mutual	Car loan	$20,500	778	Declined	Defaults in last 5 years
APP-731204	Corella Credit Union	Personal loan	$12,000	812	Approved
APP-208817	Tidewell Mutual	Debt consolidation	$34,500	702	Referred	Gambling spend share of income
APP-660093	Northgate Finance	Credit card	$8,000	745	Approved
APP-118452	Tidewell Mutual	Personal loan	$6,500	566	Declined	Credit score
APP-947310	Corella Credit Union	Car loan	$41,000	791	Approved
APP-385526	Tidewell Mutual	Personal loan	$15,000	734	Referred	Sanctions and PEP screen
APP-502781	Northgate Finance	Debt consolidation	$27,500	688	Declined	Monthly surplus at assessment rate

Whenever software applies rules, someone eventually asks why. Recount records the answer while it happens.

Pricing

Why did this customer get that rate?

Claims

Why was this claim declined?

What did the model actually see?

Billing

How was this charge calculated?

Access

Why was this request blocked?

Compliance

Can you show the auditor how?

01/How it works

A few lines around the logic you already have.

Recount is an SDK for Node.js and Python. It records what your code decided and why, in a form people outside engineering can follow.

01
Wrap the decision
Put a trace around the code that decides - a quote, an approval, a model call. Your logic stays as it is.
02
Record what mattered
Add a span for each step: the data you fetched, the rules you applied, what the model said. Flag anything unusual with an alert. Fields you mark sensitive are removed before anything leaves your app.
03
Let anyone read it
Every trace becomes a step-by-step timeline in the dashboard - for your team, for a customer in their own portal, as a row in a searchable record set, or rolled up into a report.

Full guide

quote.ts

import recount from '@refactorlabs/recount';

await recount.startTrace('Home insurance quote', async () => {
  const res = await recount.fetch('Property data', propertyUrl);
  const property = await res.json();

  const risk = scoreRiskFactors(property);
  recount.addSpan('json', 'Risk factors', { output: risk });
  if (risk.floodZone === 'HIGH') {
    recount.addAlert('Elevated flood risk', 'warn', {
      checks: [{ target: 'Flood loading', value: risk.floodLoading }]
    });
  }

  const quote = buildQuote(risk);
  recount.addSpan('json', 'Final quote', { output: quote });
  return quote;
});

quote.py

import recount

async def home_insurance_quote():
    res = await recount.fetch("Property data", property_url)
    property_data = res.json()

    risk = score_risk_factors(property_data)
    recount.add_span("json", "Risk factors", {"output": risk})
    if risk["flood_zone"] == "HIGH":
        recount.add_alert("Elevated flood risk", "warn", {
            "checks": [{"target": "Flood loading", "value": risk["flood_loading"]}]
        })

    quote = build_quote(risk)
    recount.add_span("json", "Final quote", {"output": quote})
    return quote

await recount.start_trace("Home insurance quote", home_insurance_quote)

Nine span types, each drawn the way a person would want to read it. Drop-in wrappers for fetch, MongoDB, Anthropic and OpenAI.

HTTP
Database
Table
Chart
JSON
Message
LLM
Image
Group

02/Who reads it

One record. Everyone who needs to see it.

The same trace answers your engineers, your customers and your auditors - each in the view that suits them.

Your team

All plans

The dashboard

Search traces by customer, outcome or any metadata, and step through exactly what happened. Invite anyone outside engineering - support, legal, the CEO - as a read-only viewer.

Projects for each service
Read-only stakeholder seats
Warn and error alerts on any step
Compare any two runs

Your customers

Standard & Pro

Branded portals

Give each business you serve a portal in its own logo and colours, showing only its own data, so it can answer its own “why” without opening a ticket.

Scoped to one customer’s data
Their logo and colours
Records they can search and export, on Pro
On your own domain with Pro

Your auditors

Standard & Pro

Records and reports

Turn a period of traces into a short PDF report, for your team or shared in a customer’s portal. On Pro, every decision is also kept as a record you can search and export. Every report and export is fingerprinted, so anyone can check a copy is genuine.

Four ready-made report templates
Record sets and exports on Pro
Kept long after traces expire
Emailed automatically on Pro

Your AI assistant

Pro

Or just ask.

Connect any MCP client and ask why a quote, price or decision came out the way it did. It reads the trace and explains it in plain English. Access is read-only, so it can never change your data.

search_traces
get_trace
get_daily_summary
list_projects
list_environments

AI assistantrecount connected

Why did quote QT-20840 come out at $4,200? The customer says that's high for a brick house.

get_trace · Home insurance quote

Two loadings drove it up: the property is in a high flood zone (+18%) and was built in 1985 (+12%). Brick construction (-5%) only partly offset them, and every eligible discount was applied.

03/Built for production

Safe to put in front of real data.

Recount sits in the code paths that matter most, so it’s built to be careful with what it sees.

Redaction

Mark a field sensitive and it’s removed inside your app, before anything is sent.

Encryption

Everything is encrypted in transit and at rest, and records can’t be edited once they’re made.

Retention

Choose how long each trace is kept, up to 365 days. Reports are kept separately: up to a year on Standard, or as long as you need on Pro.

Alerts

Flag any step as a warning or error, along with the values that caused it.

AI calls

Wrap your Anthropic or OpenAI client once to record every prompt, response and token count.

SDKs

Node.js and Python, with the same concepts in both. Free to start, no credit card.

04/Pricing

Start free. Upgrade when it matters.

Every plan includes the dashboard, alerts, redaction and both SDKs. Prices in AUD.

Free

For trying it out

$0/ month

50,000 spans / month
7-day retention
2 projects
2 members

Get started free

Standard

Pro

For agencies and teams at scale

$99/ month

3,000,000 spans / month
365-day retention
Unlimited projects
Unlimited members
Customer portals$9 per portal / month
Custom portal domainServe every portal from your own domain
Verifiable reportsEmailed automatically every week or month, kept as long as you need
Record setsSearchable records and verifiable exports, kept up to 25 months
MCP connector

Get started

Need higher volumes, invoicing or help with a security review? Talk to us.

05/FAQ

Questions, answered.

What is Recount?

Recount is a decision tracing SDK for Node.js and Python. Wrap the code behind a price, approval or AI answer, and Recount records every step - the data it used, the rules it applied, the model calls it made - as a timeline that your support team, your customers and your auditors can read without opening the code.

How is it different from logging, APM or OpenTelemetry?

Those tools show engineers how a system is running: what ran, how long it took and where it failed. Recount records why your software reached a particular result - the data, rules and model calls behind it - in a form your support team, customers and auditors can follow. It works alongside them rather than replacing them.

How much work is it to add Recount?

Usually a few lines. Install the Node.js or Python SDK, wrap the code behind a decision and add a step for each part that matters. HTTP requests, MongoDB queries and Anthropic and OpenAI calls are recorded automatically. Most teams start with one decision, such as quotes or claims, and add more from there.

How do customer portals work?

If you serve other businesses, you can give each of them a read-only portal showing only their own decisions, records and reports, in their logo and colours. Portals are on the Standard and Pro plans, and on Pro they can be served from your own domain, like portals.yourcompany.com.

What can we give auditors?

On the Standard and Pro plans you can generate PDF reports for any period, such as a monthly health summary or an exceptions log. Pro adds record sets, which keep every decision as a searchable row for up to 25 months, plus scheduled reports and exports. Every report and export is fingerprinted, so anyone can check a copy is genuine.

How do you keep our data safe?

Fields you mark as sensitive are removed inside your app, so they never reach us. Everything else is encrypted in transit and at rest, stored in Australia, and can’t be edited once it’s recorded. You choose how long it’s kept and who can see it. If you need a security questionnaire completed, email support@refactor.gg.

Is there a free plan?

Yes. The free plan includes 50,000 spans a month, 7-day retention, 2 projects and 2 team members. No credit card is needed.

Your software already makes the decisions. Now it can explain them.

Free to start. No credit card required.

Get started free Read the docs

Node.js guide →Python guide →