AI at Investboard
Ask a question about your own wealth. The answer is computed from your depots, measured against the mandate you signed, and written in plain language. Investboard's AI explains and checks. It never trades, never predicts, and never decides for you.

The question box sits under your plan score on Home. Type what you want to understand and press Enter. Investboard AI opens a new conversation and answers from your own data: all your depots, valued in your currency, and the plan you signed.
Home
I own 40 holdings. How many independent bets am I actually making?
Illustrative portfolio
40 holdings across cloud, semiconductors, utilities and other sectors.
The constraint
Keep the intended equity allocation.
The evidence boundary
A shared earnings driver is a hypothesis. Correlation needs return data.
Your conversation
I own 40 holdings. How many independent bets am I actually making?
Your sector labels suggest diversification. But your cloud, semiconductor and utility holdings could depend on the same AI spending cycle. The hidden concentration may be a shared earnings driver.
Shared narrative or actual correlation?
Those need separate checks: overlapping holdings, return correlations over different periods, and whether they fell together during selloffs. Holdings alone can establish overlap. They cannot establish correlation.
Which changes would reduce that shared exposure without sacrificing my intended equity allocation?
Possible approaches include smaller holdings, less overlap between funds, and exposure to different earnings drivers. Each changes the portfolio differently. The useful comparison is how much concentration each removes, what new risks it introduces, and whether it fits your mandate. That comparison needs calculated results; it is not established here.
Identify the hidden exposure
Look beyond the number of holdings and sector labels.
Challenge the explanation
Separate a plausible narrative from measured correlation.
Examine the trade-offs
Compare risk reduction, new risks and your own constraints.
Every figure in an answer was computed by Investboard before the model wrote a sentence. The answer names the rule from your mandate it measured against and the date of the data it read. What it cannot see, it says so.
Calculate first. Phrase second.
When you ask, Investboard AI does not guess. It reads your data through fixed tools: the portfolio summary, single holdings, net worth, prices as of the last close, allocation by asset class and region, drift against your plan, your tax summary, the market overview, the signals active right now, and the investment thesis you recorded. Each tool returns figures Investboard has already computed. The model may phrase them; it cannot change them, and a figure that was never computed has no place in an answer.
Target ranges, the issuer limit, the cash floor, your exclusions: an answer measures against the rules you signed, not against a general idea of a good portfolio.
Several depots are one portfolio. The answer values them together, in your currency.
Prices are as of the last market close, and the answer says so.
Elsewhere, AI works the same way, quietly: it writes your quarterly letter from figures it never sees, explains a watchlist move from the data on the card, and proposes column roles for an import file in a format we do not know yet.
03 · The boundary
Investboard AI has no path to a broker. It cannot place, change or transmit an order, and neither can Investboard.
Ask "should I buy X" and you get what the purchase would mean for your plan, never an instruction. Where an answer drifts into advice-like wording, a notice is attached to it.
Nothing in an answer predicts a price, a return or a market. Past performance is described as what it is: the past.
Every number comes from a computation Investboard ran on your data. Answers are checked against that data, and a serious violation is withheld.
Answers describe your own data and do not constitute investment advice within the meaning of § 1 (1a) sentence 2 no. 1a KWG. Any investment decision is your own responsibility. More in the liability disclaimer.
Before money moves, an intended change can be checked against your mandate: whether it fits your target ranges and limits, and which rule it touches, with the figure now and the figure it would become. Investboard describes; you decide. Today your connected AI agent runs this check (next section).
Connect the AI agent you already use. Through Investboard it can read your mandate and your saved instructions, see where your plan stands, check an intended change against your rules, and record the decision you state afterwards. With Premium and your permission, it can also read imported holdings, values and historical analytics while portfolio sharing is enabled. It cannot place an order or change your mandate. Every check remains on record in Investboard, and you can revoke access at any time.
Your mandate exports as investing.md, a readable file you choose where to share. An export connects nothing.
The connection runs through Investboard sign-in. You review the access there and can withdraw it.
Free: one connection and the core mandate tools. Premium: unlimited connections, recorded activity, your theses, your journal and portfolio analytics. Each tool needs its permission; portfolio sharing must be enabled for portfolio reads.
Connect your AI agentFree and PremiumHow to use AI in Investboard: the practical guideMCP tools and permissionsYour portable investing.md policyYour theses and decision memory
06 · Data protection
AI answers run only after you allow them, once, in the product. You can withdraw the permission in Settings; AI features then pause.
Requests run on Amazon Bedrock with a geographically bound EU inference profile, starting in Frankfurt. Investboard's client rejects any model outside that profile.
Claude, made by Anthropic, run for Investboard by Amazon Web Services. Anthropic has no access to your inputs or answers, and neither company uses them to train a model. Amazon Web Services may hold inputs and outputs briefly to detect abuse.
For each question Investboard AI receives your question, computed aggregates, the names of public securities where needed, active signals and the recent conversation. It never receives brokerage credentials, IBANs, account numbers, or your name, email and address.
They stay in your account, you can delete them at any time, and they are not used to train any model.
If you type personal data into a question, Investboard does not rewrite it. Leave out what the question does not need.
The EU processing and training commitments above describe Investboard's built-in AI. If you share a file or connect an external AI agent, that provider's own processing, retention and training terms apply. Review them before sharing. A file is a dated snapshot; it grants no access and cannot ensure model compliance.
No. It explains what your data shows and measures a change against your own rules. It cannot place an order, move money, or decide for you.
On Free, five questions a month. Premium at full capacity. Your count is in Settings.
Claude and ChatGPT, and any AI agent that supports remote MCP with sign-in. The help page has the steps.
Yes. Language models make mistakes despite the checks around them. That is why Investboard computes the figures first, keeps the decision with you, and names the data it read.
On Amazon Bedrock in the European Union, starting in Frankfurt, under an EU inference profile. AWS may distribute processing within its European geography and not outside it.
Open Mara Heller's demo and read where her plan stands. Or start with your own portfolio.