Help center › Navigator and AI

How the AI is used, and its guardrails

The AI connects, accelerates and explains; deterministic rules decide. What it does, what it never does, and how every call is recorded.

AI connects, accelerates and explains. Deterministic rules decide. An AI badge marks every text the AI wrote.

What the AI does

• Interprets requests (Intent Pipeline, what-ifs, command bars, the blotter filter) into fixed structures from a catalogue of typed actions: never code, never a database query.
• Drafts compliance and workflow rules from your words or a mandate, asks when something material is ambiguous, and quotes its sources.
• Writes explanations: why an order failed, why a proposal looks the way it does, the blotter, monitor, start-of-day, data-quality and compliance briefs, audit narratives and report drafts.
• Suggests column mappings, field definitions and missing classifications, always for you to confirm.
• Answers questions in Navigator.

What it never does

• Decide: the compliance engine, the optimizer, entitlements, workflow routing and data validation are code. A rule's restatement is generated by code from the structured rule, not by the AI.
• Act on its own: every plan is previewed and confirmed by a person; trades are proposals first; approvals and sign-offs are human, with four eyes.
• See what you cannot: it receives only data you are entitled to see.
• Change a record: data-quality issues, violations and alerts are never changed by the AI; its suggestions are staged for review.

Checks on every answer

The AI's structured answers are validated by code against the schema; invalid answers are rejected with the reasons and nothing is saved. A "passing alternative" for a failed order is shown only after the engine re-checks it. Quotes from a mandate are checked against the text.

When the AI is off or unavailable

Everything still works: explanations and drafts fall back to standard text, requests to a built-in parser, and every rule can be built with the forms. Admins switch it off under Firm settings ("Off: everything deterministic").

The record

Every AI call is recorded: purpose, model, latency, tokens, user (and API client), and the full input and output, encrypted at rest with a fingerprint of each. See AI interaction log.

Related articles

• Navigator, the AI assistant Ask questions in plain language; Navigator answers only from what you may see, and can only suggest a request for the Intent Pipeline or a command.
• The AI interaction log Every AI call: purpose, model, latency, tokens, user or API client, and the full input and output, encrypted at rest.
• Writing a rule in plain English Write a rule the way you would say it; the AI conforms it to a rule type or asks a question; code restates it; a test bench and a sign-off follow. Rules can also be extracted from a mandate.
• The Intent Pipeline Type a request in plain English; Vestoris interprets it, checks entitlements, proposes trades, pre-checks compliance, and creates orders only on your click.
• Portfolio Center commands Type or dictate what you want on Portfolio Center; Vestoris plans it, previews every change, and acts only when you confirm.
• Firm settings The thresholds that decide when a person must approve, AI use and retention, surveillance parameters and rule sign-off.

Do it with the API

Every action on this screen can also be done by your own systems, under the same permissions. See Authentication to get a token.

• POST/api/v1/copilot/questions Ask Navigator (answers only from what you may see; logged)
• POST/api/v1/commands/{screen} Plan a natural-language command and preview it (screen: portfolio, blotter, monitor, wfrules); nothing runs
• POST/api/v1/commands/{id}/execute Execute the previewed plan (X-Vestoris-Confirm: the preview token; "typed" for large or destructive plans)
• GET/api/v1/ai-log AI interactions (metadata; inputs and outputs are on the AI log page)

Paper trading only: orders and fills in Vestoris are simulated. Nothing here is investment advice.