Analytics: Agentic

Agents that act on your behalf.
You confirm before anything happens.

The platform includes agents that watch your positions, surface intelligence, and propose actions, on your behalf. When an agent reaches a consequential decision, a confirmation popup appears. You approve or reject. Without your click, nothing executes.

The principle

The agent watches.
You decide.

Ediphy Analytics agents are not autonomous. They act on your behalf, meaning they carry your identity, your permissions, and your obligations. When they surface intelligence, they assemble it for you to act on. When they want to take an action that could trigger a cost, a communication, or any irreversible step, they stop and wait for your explicit confirmation before proceeding. The scope of automation is bounded: agents handle the continuous watching and assembling; consequential decisions remain with you.

What the agents do

Four things the agent handles
so you do not have to.

Each capability runs continuously across your portfolio. The agent watches; you receive the output at the point it matters.

TRADE CONTEXT

RFQ context assembly

When an RFQ lands on the blotter, the agent assembles the market picture, fairMid comparison, recent tape, live axes, liquidity score, and delivers it before the trader's eye moves to it. Everything assembled in one view.

FAIRMID · AXES · TAPE

LIQUIDITY

Watchlist monitoring

Instruments on your watchlist that have gone quiet get flagged the moment liquidity returns. The agent watches the tape continuously; you hear about it when it matters, not when you happen to look.

WATCHLIST · ALERTS

ANOMALY

Deviation detection

Unusual moves in fairMid, large divergences from historical norms, instrument-level outliers, flagged before a manual review would catch them. The agent surfaces the signal; you decide whether to act.

OUTLIERS · NORMS

RESEARCH

Natural-language queries

Ask a plain-English question about an instrument, sector, or market condition. The agent answers from the proprietary dataset with a traceable, timestamped response, no ticket, no wait.

DATASET · TRACEABLE

How the confirmation works

When the agent wants to act,
you see this.

For any action the agent proposes, not just reads and queries, the same flow applies. The agent acts only on your click.

01 · AGENT PROPOSES

Agent identifies an action

The agent detects a condition that warrants an action on your behalf, an alert, a triggered position flag, or a workflow step. It does not execute. It queues the proposed action and surfaces it to you.

02 · POPUP APPEARS

Confirmation popup surfaces

A confirmation popup appears in your active interface, web app or mobile. It shows what the agent wants to do, in plain language: the action, the instrument, the size, the rationale. No interpretation required.

03 · YOU DECIDE

You approve or reject

One click. Approve: the action executes immediately. Reject: the agent receives a rejection event, logs it, and waits. Without your explicit approval, nothing executes. No timeout. No auto-approve.

04 · LOGGED

Full audit trail

Every proposed action, your response, and the outcome is logged with timestamp and agent identity. The audit trail is permanent and exportable, MiFID II and best-execution workflows included.

What runs freely vs what requires your confirmation
Runs automatically, no gate
Requires your confirmation, popup always
Reading positions, instruments, analytics
Any alert that triggers an external communication
Assembling trade context on RFQ arrival
Any action that creates a log entry in your name
Monitoring watchlists and tape
Any workflow step that modifies position state
Running anomaly detection across the dataset
Any action the agent flags as consequential
Answering research queries from the proprietary dataset
Any output routed to an external recipient

Why it is built this way

Agentic automation built
around the regulatory obligation,
not around it.

Every consequential action in a regulated trading environment carries a human obligation, a best-execution duty, a reporting obligation, a compliance record. Ediphy's agentic layer is built on this fact, not around it. Agents act on your behalf in the same way a junior analyst would: they do the watching, the assembling, the detecting. But they do not act without your authorisation. The popup is not a safety feature bolted on. It is the product. It is what makes agentic automation viable in a regulated context.