Enterprise: Agentic

Your agents, acting on your behalf.
Confirmation popup before any action executes.

Your agents connect to fairEx, liveBook, eContext, and the analytics APIs under their own authenticated sessions. Every read and query runs freely. When an agent reaches an action that matters, a trade, a report, a settlement, a confirmation popup is shown to the user the agent is acting for. They approve or reject. Without that approval, nothing executes.

The model

Systematic automation.
Human authority where it counts.

The Ediphy agentic API is built for one specific constraint: in regulated financial services, consequential actions have a human owner. Your agent can do the watching, the assembling, the detecting, and the proposing, but it cannot be the human who decides. The API distinguishes between actions that are safe to automate and actions that require human sign-off, and routes the second category through a confirmation popup shown to the bound human before anything executes. The popup is not optional. It is the architecture. Ediphy operates agent-driven workflows internally across its own infrastructure.

Authentication model

Agent credentials,
bound to a named human.

Every agent that connects to Ediphy operates under its own set of API credentials. Those credentials are bound to a specific named human user at provisioning time, the person the agent is acting for and the person who receives confirmation popups for gated actions.

IDENTITY

Per-agent credentials

Each agent gets its own API key pair, separate from any human user's credentials. The agent's identity is distinct and revocable independently of the human's access.

API KEYS · DISTINCT

BINDING

Bound to a named human

At provisioning, each agent credential is bound to a named human user. That human is the confirmation authority, all gated actions route to them. One human can be bound to multiple agents (e.g. a head of trading supervising a suite of strategies).

BOUND · NAMED USER

SCOPE

Permission scoping

Agent credentials are scoped to the APIs and products they need. An agent running liquidity analysis does not hold trade-submission permissions. Scope is set at provisioning and verifiable at all times.

SCOPED

AUDIT

Full activity log

Every API call made by an agent, reads, queries, proposed actions, and confirmed actions, is logged with agent identity, timestamp, and outcome. Exportable for compliance purposes.

LOGGED · EXPORTABLE

What runs freely vs what requires confirmation
Runs automatically, no gate
Requires human confirmation, popup always
Any read, query, or analytical API call
Any order submission to fairEx
Price feeds, reference data, context API calls
Any regulatory report submission (MiFIR, UK CT, EU CT)
Alerts and internal notifications
Any position or cash movement in liveBook
Research queries against the proprietary dataset
Any settlement instruction
Context assembly and anomaly detection
Any action the agent flags as consequential in its request metadata
The confirmation flow

What happens when an agent
reaches a gated action.

The flow is the same for every gated action, every time. There is no bypass. There is no timeout that auto-approves.

01 · AGENT SUBMITS

Agent submits a gated action

The agent submits the action via the API. Gated endpoints are explicit, the agent knows it is requesting something that requires human confirmation. The API accepts the request and queues the pending action.

02 · POPUP ROUTES

Confirmation popup delivered

Ediphy routes the pending action to the bound human's notification surface, web app, mobile app, or webhook to your own system. The popup shows: what the agent wants to do, the agent's identity, and the full action parameters. Nothing is summarised or abstracted.

03 · HUMAN DECIDES

Human approves or rejects

The bound human approves or rejects. One click. On approval: the action executes immediately. On rejection: the agent receives a rejection event with a timestamp. No action executes without explicit approval. No timeout. No fallback path.

04 · LOGGED

Action and decision logged

The proposed action, the human decision, the timestamp, and the outcome are all logged permanently. The audit trail is accessible to both the agent and the bound human, and exportable for any compliance purpose.

Use cases

Established deployment
patterns.

The agentic API is designed for systematic and structured workflows where automation adds value and human oversight is non-negotiable. Three patterns that teams are deploying now.

SYSTEMATIC EXECUTION

Strategy execution with human sign-off

An agent monitors market conditions and proposes trades when strategy triggers are met. A portfolio manager receives a confirmation popup showing the proposed trade, the trigger rationale, and the size. The PM approves or rejects. The strategy runs continuously; the execution decision stays with the PM.

STRATEGY · FAIRMID · FAIREX

REGULATORY REPORTING

Automated MiFIR with compliance sign-off

An agent assembles and validates the day's transaction reports. A compliance officer receives a confirmation popup showing the batch, validation results, and submission parameters. They review and approve. The assembly is automated; the regulatory submission has a named human sign-off.

MIFIR · EU CT · UK CT

BACK OFFICE

Exception detection with ops confirmation

An agent monitors settlement exceptions and proposes corrective actions, position adjustments, fails interventions, counterparty notifications. An operations manager receives a confirmation popup for each proposed correction. Exception detection is automated; the correction is human-approved before touching anything.

SETTLEMENT · LIVEBOOK

Integration

Built for your existing agent infrastructure.

The agentic API uses the same authentication layer as the rest of the Ediphy Enterprise API, your existing integration handles agent credential management. The confirmation popup can route to the Ediphy web app, the Ediphy mobile app, or a webhook endpoint in your own system if you prefer to render the confirmation in your own UI.

The JSON below shows the structure of a gated action request. The `human_confirmation_required` flag is set automatically for gated endpoints, Ediphy sets it automatically for gated endpoints.

POST /v1/agentic/fairex/order
{
"agent_id": "agt_7f2a9c", # your agent identity
"isin": "XS2345678901",
"side": "buy",
"qty": 2500000,
"limit_px": 98.41,
"rationale": "strategy trigger: liquidity score > 0.8"
}
# Response, pending human confirmation
HTTP 202 · status=pending_confirmation
{
"action_id": "act_9b3d1e",
"status": "pending_confirmation",
"routed_to": "j.smith@yourfirm.com",
"expires": null # no timeout, human decides
}_

Technical questions

What CTOs and heads of engineering ask before integrating.

Can the confirmation timeout auto-approve if the human does not respond? +

No. There is no timeout that results in execution. A pending action stays pending until the bound human approves or rejects it, or the agent withdraws it via the API. This is not configurable: the architecture does not support auto-approval.

Can we route the confirmation popup to our own system rather than the Ediphy app? +

Yes. Confirmation routing supports a webhook endpoint in your own infrastructure. Your system receives the structured pending action, renders the confirmation UI in your own interface, and calls back to Ediphy with the approval or rejection. The bound human never needs to interact with the Ediphy app directly.

What happens if an agent submits a gated action and the bound human is unavailable? +

The action stays pending. The agent can query the action status via the API, and can withdraw the pending action if conditions change. Ediphy does not substitute an alternate approver: the binding is between one agent credential and one named human.

Can multiple agents be bound to one human? +

Yes. A single human can be the confirmation authority for multiple agents, for example, a head of trading supervising several strategy agents. All gated actions from all bound agents route to the same human. The popup identifies the agent by name and ID so the human knows which agent is requesting.

Is the gated action list fixed, or can we flag additional actions as requiring confirmation? +

The standard gated endpoint list covers all trade, report, and settlement endpoints. You can additionally flag any API call as requiring confirmation by setting `human_confirmation_requested: true` in your request metadata; Ediphy will route it through the same confirmation flow even if the endpoint is not gated by default.