AI co-pilot & trust
Explainability and governance, as product surfaces
Trust is not a badge in a footer. It is a set of screens: what the co-pilot may do, how a manager reached a change, who paid for what placement, and what was logged when.
What the co-pilot does — and never does
It does
- Explains your portfolio, its changes and their estimated impact in plain language
- Shows after-cost and after-tax consequences before you act
- Compares two portfolios on risk, cost, turnover and disclosure
- Cites the version, factsheet and data it used for every answer
- Flags suitability mismatches against the profile you gave us
- Logs every answer to an audit trail you can export
It never does
- Recommend an individual security to buy or sell
- Place, modify or cancel any order
- Predict prices, returns or market direction
- Use your data to train third-party models
- Rank a manager because they paid for placement
- Answer outside its logged context without saying so
A real answer, in the co-pilot's voice
Mock conversation.
Explain my portfolio.
You hold four portfolios across three brokers, worth ₹12,45,000. 70% equity, 20% debt, 10% cash. What is working: All-Weather Core and Quality Compounders are both ahead of their benchmarks after costs, with low turnover — so little of that gain is lost to tax. What needs attention: small-cap exposure is 19% against the 12% budget implied by your stated loss capacity. Clean Energy Leaders is the largest single-theme position at 21% of equity. The timing detail: three tax lots turn long-term in 21 days, covering about ₹1.9L of gains. Trimming on the next rebalance rather than today changes the tax bracket on those lots. Sources: portfolio v7 factsheet (24 Jul 2026), your suitability profile (11 Mar 2026), broker holdings as of today. This is an explanation, not advice.
Illustrative. The co-pilot answers from your logged context and says so when it cannot.
The trust layer
Verifiable performance
Every return series is reconstructed from your actual broker transactions, net of brokerage, statutory charges, slippage and tax. Backtest and live periods are always labeled apart, and the methodology is published.
Quality-adjusted discovery
Discovery ranks on live track record, drawdown, consistency, turnover, benchmark-relative return, disclosure completeness and retention. Raw return alone never moves a portfolio up the page.
Audit logs
Portfolio versions, rationales, previews, deferrals and co-pilot answers are all logged with a timestamp and an actor. Exportable, and designed to survive a regulator reading it.
Conflict register
Every surface where money could bend a ranking, and how it is separated and disclosed.
| Surface | Basis | Paid? | Disclosure |
|---|---|---|---|
| Editorial picks | Chosen by the PortfolioOS research desk | No | Labeled 'Editorial' on every card |
| Paid placement | Manager pays for visibility in a labeled slot | Yes | Labeled 'Promoted'; excluded from ranking |
| Performance ranking | Quality-adjusted score — live record, drawdown, consistency, turnover | No | Methodology published; score components visible |
| Popularity ranking | Net inflows and retention over 12 months | No | Labeled 'Popular'; never mixed with performance |
| Sponsored research | Manager-funded factsheet or explainer | Yes | Sponsor named at the top of the document |
| Affiliated managers | Managers with a commercial or ownership relationship | Varies | Affiliation badge plus a link to the terms |
Audit trail
A sample of what gets logged, and who did it.
Portfolio v7 published
Solar +6% / legacy −6%
24 Jul 2026 · 09:14 · Meridian Research
Change rationale attached
Cited 3 sources
24 Jul 2026 · 09:14 · Meridian Research
Cashability preview generated
Per-lot tax computed
25 Jul 2026 · 18:02 · PortfolioOS
Rebalance deferred by user
Reminder set for 14 Aug
26 Jul 2026 · 10:41 · You
Co-pilot answer logged
Explain my portfolio · v7 context
01 Aug 2026 · 07:00 · AI co-pilot
Illustrative governance documentation, not legal or regulatory advice. Regulatory design must be confirmed with qualified counsel per jurisdiction. See how we compare.