🚀 Launching Soon: Get 1 Month of Growth Tier Free by signing up for Early Access

The Journey

From chaos to closed books.

Follow Order #A-4521 — a single ₹2,499 transaction — through the seven stages of a ReconPe reconciliation. Watch four agents and one deterministic engine turn a heap of files into a signed, audit-ready close.

+ ACRE Engine
Scroll to begin
Stage 01 · Arrival~3s · drag-and-drop

Files land, raw and messy.

Five spreadsheets. Two PDFs. A phone photo of a vendor receipt. Iris reads everything — including the images — without asking you to clean it first.

IrisThe Onboarder
CSVExcelPDFPhone photoJSON / XML
flipkart_apr.csv
2.4 MB
Read by Iris
amazon_settle.xlsx
1.1 MB
Read by Iris
vendor_invoice.pdf
344 KB
Read by Iris
till_receipt.jpg
1.8 MB
Read by Iris
bank_stmt.json
82 KB
Read by Iris
Ingestion zone · 5 sources captured
Stage 02 · Comprehension~6s · per source

Shape, types, structure — inferred.

Iris detects column types, the domain (Banking, Trading, Marketplace), and runs vision OCR on PDFs and images so they slot into the same stream as your CSVs.

IrisThe Onboarder

Vision pass reads scanned invoices and phone photos. The extracted rows look identical to a CSV downstream — domain-detected, type-tagged, ready for matching.

flipkart_apr.csv · 14,802 rowsDomain · Marketplace
order_id
TEXT
settle_amt
AMOUNT
settle_date
DATE
courier
TEXT
A-4521
₹2,499
2026-04-21
BlueDart
A-4522
₹4,180
2026-04-22
Delhivery
A-4523
₹879
2026-04-23
Shadowfax
vendor_invoice.pdf · vision OCR
scanning
SL-9904
₹4,499
2026-04-19
SL-9905
₹2,499
2026-04-21
Stage 03 · Unification~10s · once per source · reusable

Three names. One canonical schema.

order_id, txn_ref, ORDER_NUMBER — Iris collapses every shape into a single canonical dataset. You approve the mapping, then the matching engine never sees the mess.

IrisThe Onboarder
You approve, Iris ships.

Mapping is reviewable, editable, and reversible. The matching engine downstream only ever sees the clean canonical schema — never the messy originals.

schema mapping · drafted by Irisready to apply
source field
canonical
order_id
order_ref
txn_ref
order_ref
ORDER_NUMBER
order_ref
settle_amt
amount
Net Settlement
amount
settled_dt
settled_at
sample · row 1102 → canonical
order_ref
A-4521
amount
2499.00
settled_at
2026-04-21
Stage 04 · ACRE Match~3s · 4,103 records

Deterministic. Auditable. Zero hallucinations.

ACRE — ReconPe's matching engine — partitions records into blocks, scores candidate pairs with confidence weights, then assigns optimal matches. LLMs reason. ACRE matches.

LLMs reason. ACRE matches.

ACRE is a deterministic engine — same inputs, same matches, every run. No agent ever changes a match decision. The audit trail is reproducible from raw data.

ACRE engine · running
deterministic
source · 5
A-4521₹2,499
A-4522₹4,180
A-4523₹879
A-4524₹1,210
A-4525₹6,540
target · 5
T-1102₹2,449
VAR
T-1103₹4,180
MATCH
T-1104₹879
MATCH
T-1106₹6,540
MATCH
T-1107₹3,200
ORPHAN
matched
3
variance
1
unmatched
2
Stage 05 · Exceptions~12s · per investigation

What didn't match — and why.

Argus picks up every unmatched row and every value-mismatch. It runs an investigation pipeline and proposes resolutions with reasoning, not guesses.

ArgusThe Investigator
Reasoning, not guesses.

Argus runs a fixed investigation pipeline — match-rate, exception types, risk signals, counterparty history — and streams its reasoning live via SSE. You see why, not just what.

Argus · investigating 3 exceptionslive trace
VALUE_MISMATCHHIGH
conf 91%
A-4521 · expected ₹2,499 · received ₹2,449
whyCommission gap of ₹50 — Flipkart fixed-fee schedule
fixApply ₹50 commission tolerance for this SKU class
MISSING_TARGETMEDIUM
conf 78%
A-4524 · ₹1,210 · no settlement received
whyLikely in next settlement window (T+3)
fixDefer to next run; flag if open after 5 days
DUPLICATELOW
conf 97%
T-1107 · ₹3,200 · matched twice
whyRepeat settlement entry from rerun
fixDrop second instance
argus reasoningstep 3 of 5
→ blocking key: counterparty + amount-bucket(2400-2500)
→ candidate found: T-1102 (₹2,449), edit-distance 0 on counterparty
→ delta ₹50 = Flipkart fixed-fee for sub-₹500 SKUs
→ no prior tolerance rule — proposing tolerance addition
→ confidence: 0.91 (reason: 3 historical matches with same delta)
Stage 06 · Memoryinstant · already indexed

This month, in the context of last quarter.

Clio fingerprints every exception by counterparty + amount + direction. When this month's variance is last month's open item, ReconPe surfaces the link — one click closes both.

ClioThe Analyst
This month, in context.

Every exception gets a deterministic fingerprint — counterparty + amount + direction. When today’s surplus is yesterday’s open item, ReconPe surfaces the link before you do.

cross-run fingerprint match
fp:flipkart-x|amt:2499|dir:cr
pattern detected · 3 prior settlements
Flipkart · late settler · avg lag 5 days
auto-tagged
₹2,499 vendor variance
Raised as exception #1832 — flagged open at month-close
Feb 14
Open. Cycle moves on.
No match found. Escalated to review queue.
Feb close
Apr run
Surplus ₹2,499 surfaces
Flipkart late settlement — 67 days after invoice
ReconPe
Link & Close Both.
Fingerprint hit. Clio: "Vendor is a late settler — avg lag 5d."
Stage 07 · Closure~2s · audit bundle

Books closed. Trail intact.

Hermes orchestrates the close — disposition, approvals, audit log, downloadable report. Every decision (yours and the agents') is captured forever.

HermesThe Assistant
Hermes orchestrates the close.

Every decision — yours and the agents’ — is captured forever. Approvals, disposition, escalations, and the underlying raw evidence are bundled into one export. Auditors get the trail. You get your evening back.

SOC2-friendlyReproduciblePer-org tenancy
reconciliation closed
Run #R-204 · April Flipkart vs HDFC
audit-ready
match rate
96.1%
risk
LOW
exceptions
162
time saved
11h
audit trail · 6 entries
10:14:02IrisMapped 6 fields → canonical schema
10:14:18ACREMatched 3,941 / 4,103 records · conf ≥0.85
10:14:22ArgusSurfaced 162 exceptions · proposed 144 fixes
10:14:30ClioLinked 11 exceptions to prior runs · pattern: late-settler ×3
10:15:01amit@Approved 144 fixes · escalated 18 · signed off
10:15:04HermesClosed run #R-204 · audit bundle exported
And next month…

The cycle begins again. The memory stays.

New files. New exceptions. The same fingerprint pool, the same pattern banners, the same agents — only smarter than last month, because every decision you made taught the system one more thing.

Jan
Feb
Mar
Apr
May
→ and on
A-4521₹2,499raw row
Common questions

About the journey itself.

Bring settlement control into one audit-ready workflow.

Start with three free reconciliations each month, then scale into commission audits, COD follow-up, and shared review workflows as volume grows.