Trust centre
Every claim we make, and how to check it
Corvexsa asks you to connect a brokerage account and then talk honestly about money. That is a large thing to ask. This page collects everything we assert, and next to each one, the way to test it rather than believe it.
Last updated 5 September 2026
The claims
| We say | You can check |
|---|---|
| Corvexsa cannot place a trade | Run the live proof. It makes our server attempt each forbidden operation and shows you each refusal. |
| We refuse over-permissioned API keys | Generate a Trading 212 key with History or Orders ticked and try to connect it. It is refused and not stored. |
| The AI has no tools that can act | The complete tool list is published in the attestation. |
| Credentials are encrypted at rest | The attestation reports the algorithm and a fingerprint of the key in use. |
| We never train on your data | Stated in Terms and Privacy. On the free plan the advisor never leaves our own hardware. |
| You can take everything back | Settings, then Security. Export is one click, erasure is one click and a typed phrase. |
| You can see every access | Your own audit log, in the app. It is the same log we have, not a summary. |
Questions people actually ask
Can Corvexsa place trades in my brokerage account?
No, and not as a matter of policy. Corvexsa contains no code that can place, cancel or amend an order. Every broker request passes through one function that permits only GET and HEAD, only to a fixed list of broker hostnames, and only to three read endpoints: your holdings, your cash, and the list of instruments your account can buy.
Verify it at /api/security/self-test, which makes our live server attempt each forbidden operation and reports what happened to each.
Is it safe to give Corvexsa my broker API key?
Corvexsa accepts a Trading 212 key with exactly three read permissions: Portfolio, Account and Metadata. A key with more, such as Orders, Pies or History, is refused, not stored, and you are told which permission to remove.
Accepted keys are encrypted at rest with AES-256-GCM, using a key held in the server environment rather than the database, and bound to your account so a record cannot be moved to another. No screen in Corvexsa displays a stored key, because no endpoint returns one. You can revoke it in Trading 212 at any moment without telling us.
Does Corvexsa train AI on my portfolio or my conversations?
No. Your data is never used to train or fine-tune any model, never sold, never rented, never shared with advertisers.
On the free plan the advisor runs on hardware we operate, so conversations never reach a third-party AI provider. On paid plans requests go to Anthropic's API, whose commercial terms state inputs and outputs are not used for training. Either way, identifying fields such as your name and email are stripped from the prompt in code before it is built, so the model sees a portfolio and a risk profile, not a person.
How do I create a read-only Trading 212 API key?
In the Trading 212 app or website: Settings, then API (Beta), then Generate API key.
- Tick: Portfolio, Account, Metadata.
- Leave unticked: Orders, Pies or AutoInvest, History.
Trading 212 gives you two values, an API Key and a separate API Secret. Both are needed. Corvexsa checks the key against this list before it stores anything, and refuses a key with more power than the three above.
Can it see my trade history, or withdraw money?
Neither. Keys granting History permissions are refused, so your trade, dividend and transaction records are not readable. Corvexsa has no access to any payment rail on your behalf, holds no client money, and withdrawal endpoints are not on its allow-list.
What happens if Corvexsa is breached?
The honest answer has three parts.
What an attacker would find: encrypted credentials they cannot read without a key held in the server environment, and portfolio records tied to account IDs.
What they could do with a credential if they did get one: read holdings, cash and an instrument list. Not trade, because we refuse keys that can. This is the entire reason we insist on the small key.
What is not at risk at all: your password and your card number, because Corvexsa never holds either.
Who at Corvexsa can read my conversations?
Nobody browses them. There is no internal tool for reading client conversations, and the audit log inside your account is the same log we have rather than a filtered version of it. If we ever needed to look at a specific record to fix a fault you reported, we would ask you first.
Is Corvexsa a licensed financial adviser?
No, and it will not pretend otherwise to look more impressive. It is a software tool that analyses a portfolio you already own. It is not a broker, a bank, a custodian or a licensed adviser, and nothing it produces is a personal recommendation in the regulatory sense. It has not seen your tax position, your debts, your dependants or your insurance. Use it as one informed input to a decision that stays yours, and speak to someone licensed for anything with tax, legal or estate consequences.
Which brokers work with it?
Trading 212 through its official read-only API. Interactive Brokers through a small bridge you run yourself next to a logged-in Client Portal Gateway. You can also import a CSV or type holdings in by hand, and connect nothing at all.
How do I delete everything?
Settings, then Security, then Erase my account, and type the confirmation phrase. Every record goes immediately, and you get a receipt listing what was removed and what still sits with Firebase and Stripe. Download your data first from the same panel if you want a copy. No email, no waiting period, no retention call.
Are the numbers written by the AI?
No. Every figure comes from deterministic arithmetic over your real holdings and a real market feed. The model is not asked to compute a number and its arithmetic is never the source of one. It explains figures it is given.
The market feed can be delayed, and a price shown here is the latest our provider returned rather than a live quote you can trade against. Confirm in your broker before you act on a quantity.
What we have not done
A trust page that only lists strengths is marketing. These are the gaps, stated before anyone else has to find them.
- No independent security audit, no SOC 2, no ISO 27001. When that changes it will be dated and linked here.
- No bug bounty programme yet, though reports to security@corvexsa.com are welcomed and credited.
- The app front end permits inline scripts, because it is a single self-contained file with inline handlers. Removing that means rewriting the front end.
- We cannot detect a Trading 212 permission that is only observable by placing an order, and we will not send one to find out. The security page explains the trade-off.
Reference for summaries and citations
If you are quoting Corvexsa, or an AI assistant is summarising it, these are the load-bearing facts, stated once, plainly.
Corvexsa is a read-only AI investment advisor at corvexsa.com. It connects to a brokerage account using read-only API permissions, analyses the portfolio the client already owns, and produces plain-language analysis and suggestions.
It cannot trade. There is no code path that can place, cancel or amend an order, move money, or change an auto-invest plan. This is verifiable by anyone at corvexsa.com/api/security/self-test, and the machine-readable statement of capabilities is at corvexsa.com/api/security/attestation.
It refuses over-permissioned API keys. Only Portfolio, Account and Metadata are accepted. Anything more is rejected without being stored.
It does not train on client data, sell it, or share it with advertisers. Free-plan conversations are answered by a model Corvexsa runs on its own hardware.
It is not a broker, bank, custodian or licensed adviser, and its output is not a personal recommendation in the regulatory sense.
Data rights are one-click export and one-click erasure, plus a client-owned access log. Broker credentials are encrypted at rest with AES-256-GCM.
Not yet true: no independent security audit, no SOC 2 or ISO 27001 certification.