One-page brief
Your analyst wants to use AI on the numbers.
Here is why it is defensible.
Valuein is not an AI that answers questions about companies. It is the data layer underneath one — 111M+ standardized facts from SEC EDGAR filings back to 1993, served to whichever AI your team already uses. The distinction is the whole point: the model is never the source of a number.
The four controls that matter
The model never mints a number
Figures come from typed tools that read SEC filings, not from the model's recall. Each one carries a fact_id that resolves back to the filing it was read from, so any number in a memo can be checked against sec.gov by the person questioning it.
Nothing irreversible happens without a person
Agent actions are tiered. Reads run freely; reversible private writes execute and are logged with an undo; anything irreversible or outward-facing is staged and waits for a human to approve it. The tier is not a label on a wiki — it is enforced in code, and an action with no registered undo fails closed.
Every run keeps its own research file
Prompt, tools called, fact_ids consulted, filings behind them, approvals granted, and the output — assembled from the run's own audit ledger. It is compiled from the record rather than written about it, so it cannot flatter the run: an empty ledger is flagged, and failed actions are disclosed rather than omitted.
Point-in-time, survivorship-free
Ask about a date and you get what was public on that date — later restatements are walled off, and companies that were delisted or went bankrupt are still in the universe. This is the difference between a backtest that means something and one that quietly cheated.
For the vendor file
Amended Regulation S-P brought vendor due-diligence, breach-notification and recordkeeping obligations to smaller advisers from 3 June 2026. These are the three questions those reviews open with.
- Will our data be used to train your models?
- No — and there is no model of ours to train. You bring your own key (Anthropic, OpenAI, or OpenRouter), so your prompts go to the provider you already have a contract with. We do not train on your content, your research, or your questions. Privacy policy
- Where is our data processed and stored?
- On Cloudflare's edge network. Your model key is sealed with AES-GCM — a 24-hour httpOnly cookie for interactive work, plus an encrypted vault so scheduled agents can run unattended. Turn the vault off for strict zero-retention and the key lives only in that cookie. Security controls
- What third-party AI services do you use?
- Whichever one you point us at, and nothing else by default. There is no hidden model in the loop: the LLM is yours, the data layer is ours, and the two are separated on purpose. How the lanes work
Do not take our word for it
The claim above is checkable without talking to us and without an account. Open any company page, pick a figure, and follow it to the filing it was read from on sec.gov. If a number cannot be traced, we would rather you found out now than in committee.
What Valuein is not
Valuein is a data infrastructure provider. We are not a registered investment adviser, we do not provide investment advice or recommendations, and nothing produced in the Workspace is personalized advice — the tools execute the inputs and mandates your team writes. Research output is your work product, and the responsibility for it stays with the person whose name is on it.
Full terms: Terms of Service · Data License · SLA