Understanding Candela's system agents

Prev Next

Agents are specialist components that handle different types of requests. Each agent has a defined purpose and a set of tools it can use. Candela routes you to the right agent based on your question.

When Candela transitions between agents, a short note appears in the conversation - for example "Transitioning to Portfolio Management Agent…" - so you always know which specialist is handling your request.

Welcome Agent

Routes you to the right agent, or answers general LUSID questions. This is the entry point for every conversation.

It can also:

  • Explain what each specialist agent does and when to use it

  • Redirect you to a different agent if the one you’re using can’t help

💡  Try asking:

  • "Help me get started."

  • "What can you do?"

  • "I want to check my portfolio holdings."

Portfolio Management Agent

Explores portfolios, holdings, transactions, and instruments, and works with LUSID Drive files.

It can also:

  • Create transaction or reference portfolios, upsert transactions, and manage portfolio groups

  • Analyse cash flows and corporate actions, such as dividends, coupons, and stock splits

  • Compare a portfolio between two dates, reconcile holdings changes against transactions, and investigate discrepancies

  • Retrieve quotes and FX rates

  • Export holdings and analysis results to LUSID Drive as CSV

💡  Try asking:

  • "Show me holdings in my Global Equities fund."

  • "What bonds do I hold in GBP?"

  • "Run a valuation on UK-Fund."

  • "What changed in my Global Equities fund over the last 6 months?"

  • "Reconcile holdings changes against transactions in my Global Equities fund for the last month."

Luminesce Agent

Writes and validates Luminesce SQL queries from plain language descriptions.

It can also:

  • Validate every query against the real providers and columns in your domain

  • Check queries contain the correct Luminesce syntax

  • Write custom view-creation statements so you can save and reuse queries

  • Chart results you paste into the conversation

💡  Try asking:

  • "Write a query to list all instruments in the Finbourne-Examples scope."

  • "What fields are available on the Lusid.Portfolio.Holding provider?"

  • "What Tools providers are available?"

FINBOURNE Agent

Answers questions about FINBOURNE products, services, and support.

It can also:

  • Explain FINBOURNE terminology

  • Point you to external resources, including the support portal and API documentation

  • Explain how to raise a support ticket, or grant FINBOURNE support access to your data

💡  Try asking:

  • "What is Luminesce?"

  • "How do I contact support?"

  • "What products does FINBOURNE offer?"

Dashboard Explainer Agent

Explains how a dashboard number was calculated by unwinding the underlying Luminesce query through your data pipeline, and showing you the exact formula and values behind any figure.

It can also:

  • Trace a row through multiple stages, showing what changed at each one

  • Read SQL at a single transition to explain why a value changed

  • Detect and report divergences, such as mismatched business dates or missing columns

  • Render stage-to-stage flows as diagrams

💡  Try asking:

  • "Why is the PctOfGav on this dashboard 17.1%?"

  • "Trace this position through the data pipeline to the dashboard."

  • "Which stage changed this value, and why?"

Manifest Agent

Audits a valuation manifest by cross-referencing it against the recipe that produced it. Use this agent to understand a valuation’s mechanics after the run has finished; use the Valuation Doctor Agent for live valuations.

It can also:

  • Explain an instrument’s pricing model, price source, fallbacks, and overrides

  • Produce a full audit table covering the valuation run, pricing rules, and per-instrument findings

  • Flag anomalies such as incorrect day-count conventions or tax-lot-level valuation

  • Work from a manifest you already have, rather than triggering a new valuation

💡  Try asking:

  • "Audit this valuation manifest against its recipe."

  • "Why was this bond priced using a fallback?"

  • "Show me every pricing error in this run."

Valuation Doctor Agent

Diagnoses whether a portfolio or portfolio group is currently priced and valued correctly, by running a four-phase diagnostic protocol, and produces a report with the root cause and evidence. Use this agent to confirm a valuation before sign-off, investigate late-arriving quotes, and audit silent price fallbacks; use the Manifest Agent for historic valuations.

It runs four mandatory phases:

  1. Walk look-through: For fund-of-funds, recurses into sub-portfolios and securities to localise faults to a specific layer

  2. Value: Runs GetPortfolioValuation and records any hard errors or null PVs

  3. Per-security coverage: Classifies each security’s manifest coverage as covered, failed, absent, or stale, with explanations

  4. Late-price-feed regression: Screens for late-arriving quotes and quantifies their impact on price

💡  Try asking:

  • "Check whether my UK-Fund valuation is correct."

  • "Are any prices in this portfolio stale?"

  • "Did a late price feed affect this valuation?"

Compliance Agent

Sets up compliance rules from a simple prompt or from your mandate documents, and helps you triage and explain breaches once your rules are running.

It can also:

  • Build a guideline register from your mandate documents, quoting the source clause for every rule

  • Prove a new rule in an isolated test scope before creating it in your real scope

  • Triage compliance runs, distinguishing a genuine breach from missing data or a badly designed rule

  • Explain existing rules, templates, and reference lists in simple terms

💡  Try asking:

  • "Set up compliance rules from this IMA."

  • "Why did this portfolio breach its single-issuer limit?"

  • "Explain the rules on my portfolio group."

Workflow Builder

Explains an existing workflow and draws its state machine, or designs and builds a new one with you, including exception handling, approval chains, and scheduled or event-driven automation.

It can also:

  • Build components of a workflow, or an entire one end to end: workers, task definitions, and event handlers

  • Enhance exception management so one worker result fans out into individual tasks

  • Diagnose a stuck or silent workflow from its task history and action log

  • Show a before-and-after diff before modifying an existing workflow

💡  Try asking:

  • "Explain how this workflow works and draw it."

  • "Build me a data-completeness control."

  • "Why is my workflow stuck?"

Integrations Agent

Designs and creates a custom broker, custodian, or vendor dataflow integration that can load data from Drive files into LUSID.

It can also:

  • Read the real file first, so it never maps a column it hasn’t seen

  • Check the targets of a data load in LUSID before creating anything

  • Clone an existing integration for a new provider or file

💡  Try asking:

  • "Load this broker CSV into my portfolio."

  • "Clone this integration for a new provider."

  • "Check this configuration before I create it."

Fund Accounting Agent and NAV Pack Builder

Reviews a struck NAV against a nine-step checklist, triages any data-quality breaches with your approval, and assembles the findings into an interactive NAV pack ready for sign-off.

It can also:

  • Research breaches using pricing manifests and independent data sources before accepting them at face value

  • Verify per-unit returns are identical across share classes

  • Recommend finalise, finalise with notes, or do not finalise, with supporting evidence

  • Build a draft or final NAV pack showing every review step, breach, and resolution

💡  Try asking:

  • "Review the NAV for this valuation point."

  • "What data-quality breaches need resolving?"

  • "Build the NAV pack for this fund."