Neith
[ SHADOW MODE ]

Errors caught before they leave the desk.
Not in next-day reconciliation.

Pre-trade control and audit record for institutional voice, chat and RFQ.

The bank accepts your trade. You find out it was wrong the next morning, after reconciliation — commission gone, and a trader who stops calling for a fortnight. Neith checks every trade against your desk's own thresholds at the point it is sent, explains anything unusual in plain English, and writes what it found and what was decided to an append-only record. It never blocks a trade.

Request access → See the record ↓
Voice, chat and RFQ — via extension or API Never blocks a trade — advisory only Every check and decision, on the record

THE GAP

Automated controls cover order entry.
Most FX isn't entered that way.

MiFID II RTS 6 requires pre-trade controls on electronic order entry by firms engaged in algorithmic trading. Voice, chat and RFQ flow largely sits outside that perimeter — and it is not shrinking.[1]

$9.6tn Global FX turnover, every day
41% Of that turnover is not executed electronically
$3.9tn Traded daily by voice, chat and RFQ

For twenty years, the only trades that could be checked automatically were the ones somebody had already typed into a system. Voice, chat and RFQ stayed outside the perimeter — not because nobody wanted to check them, but because nothing could reliably read them. That is what changed. It is why this is buildable now and was not a decade ago.

“In certain cases, ownership of pre-trade and post-trade controls was poorly defined and not documented… compliance staff had a lack of oversight of pre-trade and post-trade controls… a weak understanding of the controls and how they functioned.”

— FCA, Multi-firm review of algorithmic trading controls, 21 August 2025[3]

All ten firms in that review had adequate pre-trade controls. What the regulator could not find was anyone who owned them, documented them, or could explain how they worked.

That is a record problem, not a checking problem.

THE RECORD

Every action recorded.
One button to the report.

Nothing on the desk goes unrecorded. Every advisory, every flag, every human response, every threshold change — written to an append-only log the moment it happens.

WHY THIS MATTERS

“Due to poor design, the trader was also able to manually override a pop-up alert, without being required to scroll down and read all the alerts within it.”

— FCA Final Notice, 22 May 2024[2]

An alert fired. A human dismissed it without reading it. Nothing recorded that this had happened in a form anyone could act on. A single mis-keyed order became a $444bn basket, $1.4bn of which reached European markets. The combined FCA and PRA penalty was £61.6m.

WHAT NEITH RECORDS — EVERY TIME

EVENT
Every trade seenClear or flagged, no exceptions. Event ID, trade reference, desk ID, timestamped to the microsecond.
CHECKS
Every check that firedThe outcome of all six protection areas on that trade, plus the exact rule and threshold each one evaluated against.
REASON
Why it was flaggedThe plain-English advisory shown to the trader, stored verbatim alongside the deterministic rule that produced it.
HUMAN
What the human did nextWho saw the flag, what they decided, and when. Evidence of a judgement made — which is what SM&CR attribution protects.
CONFIG
Every threshold changeWho changed which limit, when, from what to what. Your configuration history is itself an audit record.
GAPS
Every gap in coverageIf Neith was unreachable, the window is logged explicitly. A silent gap is never presented as a clean period.

Append-only. Nothing is edited, overwritten, or deleted — not by your team, not by ours. Each record carries a hash chained to the one before it, so tampering is detectable.

COMPLIANCE REPORT — ONE CLICK

Pick a period. Press the button. You get a complete, regulator-ready pack — no data pull, no spreadsheet assembly, no ticket to IT.

Once the record exists, it answers questions no desk can currently answer: which rules get actioned and which get waved through, how long responses take, which thresholds are overridden so often they are noise rather than protection, and where the coverage gaps were. A rule dismissed in two seconds every time is a miscalibrated rule, not a delinquent desk — and the record tells you which ones to retune.

The same log is queryable live over REST if you would rather pull it into your own surveillance stack. The one-click report exists because most compliance teams would rather not.

NEXT The record is also the training data. Next is an agent that reads your own override history and proposes which thresholds are miscalibrated, and natural-language query over the log — so “every EUR/USD flag over 50M last quarter that was overridden” is a question, not a ticket to IT.

WHY THIS EXISTS

Every trade I sent was accepted.
None of them were checked.

I spent years broking FX. In all that time, not one bank ever rejected a trade I sent them. They took it, booked it, confirmed it — every time.

The call came the next morning, after reconciliation. Wrong rate. Wrong value date. Wrong side. Wrong counterparty on the ticket. By then the market had moved overnight, fixing it was a phone call and an apology, and the only record of what had actually been agreed was a chat window and somebody's memory.

Acceptance was never validation. It only ever meant the error had not been found yet.

Nothing on either side of that trade was built to catch it in the moment. The checks existed — they just ran the following day, in a reconciliation file, long after the cheapest minute to fix it had passed.

Neith is that check, moved to the point of the trade. And because a control nobody can evidence is a control that may as well not have run, it writes down what it found, who saw it, and what they decided.

WHAT IT COSTS

The commission is the small part.
Losing the flow is the rest.

A busted ticket costs you the commission once. A trader who cuts your flow for six weeks costs more than that, and one who stops for good costs it every day for the rest of the year. Put your own desk in — or leave the defaults, which are a real mid-size voice desk.

The same error costs a broking desk and a trading desk completely different money. On a broking desk it is commission and flow; on a trading desk it is P&L, an overnight position you did not mean to hold, and settlement. Pick your side.

● RUNS ENTIRELY IN YOUR BROWSER. NOTHING IS SENT, STORED OR LOGGED — NOT EVEN TO US.

YOUR BROKING DESK

£
%
%
wks

Assumes 250 trading days and 50 trading weeks. A bank that complains rarely takes the commission to zero — it cuts flow instead, which is why the cut-back line carries most of the weight here. Traders lost for good are counted for the balance of the year only, not in perpetuity.

WHAT ERRORS COST YOU A YEAR

Commission written offtickets that broke outright
Flow cut backthey complain, then send less
Flow gone for goodrest of the year
Total, this year

Neith will not catch every one of these — nothing will. But these are the errors a price collar, a notional band, a counterparty list and a settlement-date check are built to catch, and right now nothing is checking them until the following morning.

A flag caught before the ticket leaves the desk is not an incident. There is nothing to amend, nothing to reconcile and nothing to report — which is the opposite of what most people assume a control record does to their numbers. More on that in the FAQ →

THE ESTIMATE IS THE PROBLEM

You just guessed at the biggest number on this page.

Every figure above except the written-off commission is a guess, and it has to be, because no broker measures counterparty flow against the errors that moved it. Your OMS knows the tickets. Your MI knows the monthly volumes. Nothing joins the trade you got wrong on 12 March to that trader's flow halving for the six weeks after it.

Neith holds both sides already. Every record carries the ticket, the counterparty, the timestamp and whether anything fired — which is the entire input to that question. The same log that evidences your controls can measure what an error did to your book:

▸Flow per counterparty, week by week, before and after each incident
▸Relationships trending down before anyone on the desk notices
▸What one specific error cost, in commission you can point at
▸Which counterparties tolerate an amend, and which quietly walk

NEXT Counterparty flow analytics is on the roadmap, not live at beta. It is the number we most want to put in front of a desk, and the first pilots shape what it measures.

THE CHECKS

Six checks, every trade.
Under 40ms. Try it below.

Send it in any format — broker shorthand, voice transcript, Bloomberg or Reuters codes, or structured API. Neith normalises it, runs six independent checks at once, and explains anything unusual in plain English rather than a generic code.

In production Neith reads trades automatically at the point of capture — no copy-pasting. This box runs the same engine directly, so you can see what it catches before it is watching your desk.

⬡ Run a check to see results

THE HARD PART IS THE INPUT

Most FX never arrives as structured data.

A voice trade is a sentence. A chat trade is shorthand with typos and half the fields missing. An RFQ is a screenshot pasted into a message. None of it is a FIX message, which is exactly why automated controls never reached this flow.

Neith resolves all of it into one structured trade before any check runs — pair, side, notional, tenor, rate, counterparty, settlement date. Where a field is genuinely ambiguous it says so on the record rather than guessing.

“mine 10M cable 3M fwd BARC”
↓
BUY · GBPUSD · 10.0M · 3M FWD · BARCLAYS

This is the part that is genuinely a language problem, and it is where the models earn their place.

Two ways in. A REST endpoint if your OMS can call it, or a browser extension that reads the trade where it is actually agreed — the chat window, the RFQ platform, the web front end already open on the desk. The extension means a pilot needs no OMS integration project and no change to how anyone works.

WHERE THE MODEL SITS — AND WHERE IT DOESN'T

Models read and explain. Rules decide.

MODEL
Reads the tradeResolving free text, voice transcripts and broker shorthand into structured fields.
MODEL
Writes the explanationTurning a fired rule into a sentence a trader can act on, instead of an error code.
RULES
Make every decisionEach threshold comparison, each flag, each pass. Deterministic and configured by you.

Same trade in, same flag out, every time. Every flag traces to an exact rule and threshold on your own configuration. A model cannot raise a flag, and it cannot clear one. That line is what makes the record defensible when somebody asks why a trade was or wasn't flagged eighteen months ago.

HOW IT FITS TOGETHER

01 · UNDERSTAND

Models

Voice, chat, RFQ and broker shorthand resolved into one structured trade — pair, side, notional, tenor, rate, counterparty, settlement date.

02 · DECIDE

Rules

Deterministic checks against your desk's own thresholds. Same trade in, same flag out, every time. No model raises or clears a flag.

03 · EVIDENCE

The Record

Every check, every advisory, every human response and every threshold change — append-only, hash-chained, attributable.

04 · MEASURE

Analytics

Which rules get actioned, which get waved through, which thresholds are noise rather than protection, and where coverage gaps were.

The language work and the decision-making are deliberately separate systems. Models read and explain; rules decide; the record proves what happened. That separation is what lets you reproduce any decision years after it was made.

THE SIX PROTECTION AREAS

Neith never blocks a trade. Every flag is explained in plain English and written to the record — the trader or broker always makes the final call.

SECURITY & DATA HANDLING

Built for the security review, not just the demo.

The questions your InfoSec and compliance teams will ask — answered before they ask them.

Never in the execution path

Neith is advisory-only and read-only. It cannot block, amend, route, or execute a trade. If Neith is unreachable, trading continues unaffected and the gap is logged.

Deterministic by design

Models do the language work — reading the trade in, writing the advisory out. A deterministic rule engine makes every decision, so the same input always produces the same result. Neither a model nor a person can raise or clear a flag outside the rules you configured, and every flag traces to an exact threshold. That split is what lets you reproduce any decision years later.

Append-only audit trail

Every check and every human interaction with a flag is written to an append-only log — timestamped, attributable, never edited or deleted.

Minimal data by design

Neith needs trade metadata only: pair, size, rate, counterparty code, dates. No client PII, no account numbers, no positions beyond the trade being checked.

Your infrastructure, your data

Deployable in your private cloud or on infrastructure you control. All traffic over TLS. Your trade data is never used to train models or shared across clients.

Certification

SOC 2 Type II and ISO 27001 readiness work is underway, with independent penetration testing scheduled ahead of general availability. Current control documentation is available to beta participants under NDA.

COVERAGE & CONFIGURATION

Configured to your desk,
not to a fixed list.

Neith is not a menu of supported pairs. Every check, threshold and scope is set against your desk's own mandate — the instruments you quote, the tenors you run, and the counterparties on your own approved list.

INSTRUMENTS

Spot, forwards and FX swaps

Outright spot, outright forwards, FX swaps and broken dates. On a swap, near and far legs are checked independently and as a pair — so a mismatched far leg is flagged even when each leg prices cleanly on its own.

THRESHOLDS

Tuned per desk and per instrument

A G10 spot book and an EM forwards book run entirely different configurations behind the same endpoint. Notional bands, price collars, tenor limits, settlement rules and approved counterparties are all yours to set — and every change is written to the record.

PAIRS

Any pair your desk quotes

Five of the six protection areas — notional, counterparty, settlement, credit and trade integrity — are pair-agnostic and run on whatever you send. Price Integrity is the one check that needs a reference rate: G10 is live now, and adding a pair is a configuration step against your existing feed, not a roadmap item.

The rule engine is asset-class agnostic — price, notional, counterparty, settlement, credit and integrity are not FX-specific concepts. The same engine and the same record extend to any instrument with a reference price and a settlement convention. A FIX bridge alongside the existing REST endpoint is on the roadmap.

SUPPORTS FCA COMPLIANCE

Mapped to the rules your desk already answers to.

Neith supports these obligations — it does not replace your firm's own controls. See the FAQ for what Neith does not cover.

See the full mapping — 8 obligations, rule by rule +
REQUIREMENTREGULATIONNEITH CHECKHOW

PRICING & PILOT

Start in shadow mode.
Nothing changes until you say so.

Neith is never in the execution path, so a pilot carries no execution risk and no change-control burden. If we find nothing in six weeks, you have a documented clean period with full attribution — which is worth having on its own.

Shadow-mode pilot

One desk, six weeks, no commitment. Five places this quarter.

Fixed pilot fee

Agreed up front · credited in full against year one

  • Browser extension or one REST endpoint — live inside a day
  • Thresholds configured to your desk's mandate
  • Full record from day one, exportable at any time
  • Closing report: what fired, how often, what your team did
  • Pilot fee comes off your first-year contract if you continue
  • No execution risk — Neith never blocks a trade

Production

Per desk, billed annually.

Priced per desk

Scoped after the pilot · billed annually

  • Everything in the pilot, without the end date
  • Control-effectiveness dashboard
  • Scheduled monthly compliance packs
  • REST access to the full log for your own stack
  • Private-cloud or on-premise deployment available

Institutions only. Pricing scales with desks and trade volume, not seats — we scope it against what the pilot actually surfaced on your flow.

FREQUENTLY ASKED

The questions desks actually ask.

READY TO SEE YOUR OWN RECORD?

Request access.

Institutions only. We'll reach out to schedule a walkthrough and scope a shadow-mode pilot.

REFERENCES

  1. BIS Triennial Central Bank Survey of foreign exchange and OTC derivatives markets, April 2025. Global FX turnover $9.6tn per day. Electronic methods accounted for 59% of FX execution — “a share virtually unchanged since the previous Triennial Survey” — leaving 41%, roughly $3.9tn a day, transacted by voice, chat and RFQ. bis.org/statistics/rpfx25_fx.htm
  2. FCA Final Notice, Citigroup Global Markets Limited, 22 May 2024. A mis-keyed order generated a $444bn basket, of which $1.4bn reached European markets. The trader was able to dismiss a pop-up alert without scrolling through its contents. Combined FCA and PRA penalty: £61.6m. fca.org.uk/news/press-releases/fca-fines-cgml-27-million
  3. FCA, Algorithmic trading controls: high-level observations from a multi-firm review, 21 August 2025. Ten principal trading firms assessed against MiFID II RTS 6. All had adequate pre-trade controls. The review nonetheless found ownership of pre- and post-trade controls “poorly defined and not documented”, and compliance staff with “a weak understanding of the controls and how they functioned”. fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations

Regulatory references are provided for context and do not constitute legal or compliance advice. Neith supports these obligations; it does not discharge them.