AI & Agents Terms

The Workspace chat assistant ("Valuein AI" when it runs on Valuein's managed model, or the provider you chose when it runs on your own key), Agents, briefs, and tools that draft theses, claims and reports are AI Features. Their output is generated by a language model. It is not written or reviewed by a person at Valuein, and it is not written by a licensed professional. Every public research page notes that AI may have drafted parts of it.

Language models can misread, misstate, omit or invent information, including figures, dates, names and citations, and can do so confidently. Valuein's data tools are built so that a model does not need to produce a number itself: each figure comes back from a tool with a fact_id, a source filing, a unit and a period, and Valuein's models are instructed to quote those values exactly and never invent one. That reduces errors; it does not remove them. A model can still copy a figure wrongly, attach it to the wrong company or period, draw a wrong conclusion, or write a number that no tool returned. A figure with no fact_id link did not come from Valuein's data tools.

Before you rely on AI Output, publish it, or give it to anyone, review it and trace the figures that matter to their filings (in the Workspace, follow the link; in the MCP Server, call verify_fact_lineage). AI Output is provided "as is" under Terms §19, is not investment advice (see the Financial Disclaimer), and is your work product once you use it.

You can use Valuein Data with models Valuein does not operate: your own key in the Workspace (Anthropic, OpenAI or OpenRouter), an MCP client such as a chat assistant or IDE that you connect to the MCP Server, the SDK or Bulk API inside your own code, or any Agent you build.

In each case Valuein returns deterministic data through typed tools and documented endpoints. What another model does with that data — how it interprets, summarises, combines, rounds or misstates it — is outside Valuein's control, and Valuein is not responsible for it. This is a disclaimer of responsibility for another model's output; it does not make you indemnify Valuein for it, and it does not limit Valuein's responsibility for the data its own tools returned, as set out in the Terms.

An Agent is any automation that uses the Service through your Account or Credentials, including Valuein's own Agents running for you and Signals or schedules that start a run without you being present.

• You are responsible for how each Agent is configured (its goal, tools, model, schedule, Signals and memory), for the Credentials and model keys it uses, for every action you approve, and for everything it produces or does under your Account, as if you had done it yourself.
• Which Agents these controls cover. The approval, staging and labelling controls below apply to Agents that Valuein runs: the Workspace chat, Workspace Agents and scheduled runs. An AI client you connect yourself (for example, a desktop assistant using your MCP connection or API key) calls Valuein's tools directly; Valuein cannot see or supervise its reasoning, and the actions it takes with your Credential are yours.
• Approvals are aids, not guarantees. Valuein classifies Agent actions by risk and stages higher-impact actions (for example, sending anything outward, or publishing) for your explicit approval, and records approvals and rejections in an audit log. These controls help you supervise; they are not a promise that every risky action will be caught, and approving an action makes it yours.
• No trading, no money movement. Valuein's Agents cannot place trades, send orders or move money. They produce research and manage your Workspace records.
• Publishing needs you. An Agent that Valuein runs cannot publish a thesis, claim, report or profile item to the public on its own: an Agent-initiated publish is staged until you approve it, and every public research page notes that AI may have drafted parts of it.
• Limits apply. Agent runs count against the same Plan limits, quotas and charges as your own use. Valuein may throttle, pause or disable an Agent that loops, malfunctions, breaks the Acceptable Use Policy or degrades the Service, and will tell you when it does.

Seat — one named individual human. A Seat may not be shared. Agents may act only for the single human Seat holder who controls them; an Agent that serves, or whose output is delivered to, any other person, team, firm or client needs a Seat for each person it serves or a Commercial Agreement.

AI Features read text that neither you nor Valuein wrote: SEC filings and exhibits, documents you upload, other users' Published Content, and web pages or tool results from connected services. Any of it can contain instructions written to manipulate a model — to make it ignore its rules, reveal information, misreport figures or take actions you did not ask for.

Valuein applies defences, including marking third-party text as untrusted for the model and limiting what actions an Agent can take without your approval. No defence is complete. Treat surprising AI Output, especially a proposed action you did not request, as a possible injection, and do not approve it without checking. Do not upload documents you do not trust, and do not connect an Agent to systems or Credentials it does not need. Deliberately attacking the Service this way is prohibited by the Acceptable Use Policy.

The Acceptable Use Policy applies in full to AI Features and Agents. In particular, you must not use them to manipulate markets, tout or scalp securities, mass-produce misleading financial content, impersonate anyone, extract Valuein's system prompts or other users' data, circumvent Plan limits, or train models on Valuein Data except as the Data & API License Schedule permits.

We do not use your prompts, uploads, saved research, Agent memory or AI Output to train, fine-tune or evaluate any machine-learning model.

If you send us Feedback (for example, reporting a wrong figure), we may use it to fix and test the Service — for instance, turning a confirmed data error into a regression check of the filing value behind a fact_id. That check contains the filing's value, not your prompt or AI Output.

Every AI Feature runs on one of two lanes. The lane is shown in the Workspace when you choose it.

Your key. Valuein never stores your key in plaintext. It is sealed with AES-GCM; Valuein keeps the sealed copy in a browser cookie that expires within 24 hours and also in an encrypted key vault so your standing automations can run. The key is decrypted only in memory, only to call your provider. You can remove the vaulted copy at any time with the vault switch in chat settings, or remove the key entirely; removal applies to future runs immediately. Removing the vaulted copy does not change how your provider retains content.

Outside the Workspace. When you use the MCP Server, SDK or Bulk API from your own tools, Valuein only returns data; the model and client you use are your choice and process that data under your terms with them.

The full list of providers is in the Privacy Policy and on the Subprocessors page.

Valuein's data tools are deterministic: the same tool, the same inputs and the same data snapshot return the same result, including point-in-time (as_of_date) queries. A new data release, an issuer's restatement or a correction can change results between snapshots. Model output is not deterministic: the same prompt can produce different wording, different tool calls and different conclusions on different runs, and model providers change their models over time. Valuein does not promise that two AI runs will produce the same output.

AI Features change often. Valuein may add, change or remove models, providers and AI Features under Terms §23. A change in who processes your content, or any change to section 8, is a material change under the Terms and is notified 30 days in advance.