Ask anything about your client's portfolio.

"What's the Johnson household's current allocation to alternatives?"

"Show me all positions with unrealized losses over $25,000 that have been held more than a year."

"How much did we bill the Chen household in Q4, and what was the effective rate?"

"What's the total unfunded commitment across all private equity funds for this household?"

"Compare YTD performance against the blended benchmark — which asset classes are dragging?"

"Are there any capital calls due in the next 60 days?"

These aren't hypothetical examples. These are the questions advisors ask every day in client meetings, annual reviews, and portfolio check-ins. Advisd's AI answers them instantly, using the household's actual data.

Classification. A two-layer system determines what data the question needs. The first layer uses rule-based pattern matching for common question types (allocation questions, performance questions, billing questions). If the pattern layer doesn't match, a lightweight LLM classifier categorizes the question and identifies which data sections are relevant.

Context assembly. The engine builds a data context specific to the question. Rather than dumping raw database rows into the prompt, Advisd generates computed section summaries — liquidity rollups, allocation breakdowns with cross-references, fee calculation details, performance comparisons. This produces more accurate answers and keeps the AI focused on what matters.

A baseline context covering the household's core profile, account structure, and high-level allocation is always included. This handles roughly 80% of questions without any additional data retrieval.

Response generation. The AI generates a response grounded in the assembled data. Answers cite specific numbers from the portfolio. When the question involves calculations — like comparing performance across time periods or aggregating values across accounts — the response shows its work.

Guardrails. Every response passes through compliance checks. The AI never provides investment recommendations, buy/sell advice, or forward-looking projections. Responses include mandatory disclaimers appropriate to the audience. All queries and responses are logged with timestamps, user identity, and the data context used.

Advisd's AI is audience-aware from the first step. The system knows whether the question is coming from an advisor or a client, and adjusts the depth, language, and guardrails accordingly.

Advisors get full analytical depth. Responses include granular position-level data, fee calculation details, tax lot specifics, unrealized gain/loss breakdowns, and cross-references between modules. Advisors can ask about billing internals, trading history, and compliance status.

Clients get factual answers in accessible language. The client portal's AI query bar answers questions about their portfolio — "What's my current allocation?" "How has my portfolio performed this year?" "What fees have I been charged?" — using the same underlying data, but filtered for client-appropriate access. No internal billing details. No advisory content. No recommendations. Mandatory disclaimers on every response.

This isn't two different systems. It's one AI engine with two audience modes, ensuring that clients get accurate information without crossing regulatory boundaries.

A general-purpose AI chatbot doesn't know your clients. It doesn't know the difference between a capital call and a margin call. It can't tell you that the Smith household's alternatives allocation is 18% because it doesn't have access to the Smith household's data. And when it doesn't know, it guesses — confidently.

Advisd's AI intelligence layer is fundamentally different.

Domain-specific. The AI understands RIA concepts natively: households, AUM, TWR vs. MWRR, capital calls, breakpoint tiers, wash sale rules, suitability profiles. It doesn't need these concepts explained in the prompt.

Data-grounded. Every answer is computed from the household's actual data. When the AI says "your alternatives allocation is 18.3%," that number comes from a real calculation across real positions — not a statistical guess.

Contextual. The AI sees the full picture for each household: positions across accounts, performance history, fee schedules, alternatives commitments, tax lots, and activity timeline. It can cross-reference between these — "your alternatives allocation increased by 3% since last quarter because of the $500K capital call to Fund X."

Constrained. The AI is designed to tell you what is, not what should be. It answers analytical questions. It does not provide investment recommendations, predict market movements, or suggest portfolio changes. These constraints are architectural, not just prompt instructions.

AI in a regulated environment requires more than good intentions. Advisd's AI intelligence includes compliance controls at every layer.

No investment recommendations. The system is architecturally constrained to answer questions about existing portfolio data. It does not suggest trades, recommend allocations, predict performance, or provide any content that could be construed as investment advice.

Mandatory disclaimers. Every AI response includes appropriate disclaimers. Advisor-facing responses note that the AI provides informational analysis only. Client-facing responses include regulatory disclosures and encourage clients to speak with their advisor.

Full audit trail. Every query, response, classification decision, and data context is logged with timestamps and user identity. These logs are retention-protected and available for compliance review and regulatory examination.

Rate limiting and usage tracking. Query volume is monitored per user and per organization. Usage data is available to firm administrators for oversight and cost management.

Suggested query chips. Rather than leaving users to wonder what they can ask, the interface presents contextual query suggestions based on the household's available data. These are curated, factual question types — not prompts that could lead toward advisory territory.