Security, control and your data.
Crevox is being built around permissions, approval rules and reviewable work. Here is what those controls mean, how AI providers fit into the data flow and what your team should evaluate before connecting company information.
On this page
Access and approval are different controls.#
Tool access determines which supported services an agent can use. Approval rules determine whether a proposed action needs your review before proceeding. Crevox is being designed to let you configure an agent’s independence and choose which routines can run on their own.
For a customer follow-up, your workflow might allow preparation of a draft while keeping the final action behind an approval. The available actions depend on the connection and the permissions you grant.
Turning off a connection stops that connection. It does not remove information already shared with a provider or undo an action already completed.
Where does company context go?#
An AI task can involve Crevox, the selected model provider and the tools you connect. The provider needs the task context used to generate its response. Connected services process information according to their own policies and the actions you request.
The planned Mac option changes where Crevox runs. It does not make a cloud model local or prevent the model provider from processing task context. Choose the account and connection whose policies fit the information you intend to share.
The separate waitlist privacy policy explains the email, referral and optional survey data collected by this website.
Training settings depend on your account.#
Personal ChatGPT accounts provide an “Improve the model for everyone” control. Codex environment sharing has a separate setting; some features and feedback have additional rules. Review the settings for the actual account and connection you use. Read OpenAI’s data controls.
Grok provides an “Improve the Model” control for new conversations on its app and website. Feedback and using Grok through X have separate considerations. Read Grok’s data controls.
OpenAI’s business and API products exclude business data from model training by default unless you opt in. That policy is separate from processing and retention. Read OpenAI’s business data policy. Turning off training is not a promise that task data is never processed or stored.
Check the source and the result.#
Crevox’s planned knowledge workflow links answers back to saved sources and lets you correct company knowledge. Task and routine views let you follow a run and review its result.
AI can make mistakes. Check that a source is relevant and current, and that the proposed result matches the task. Keep human review around work that affects customers or important business decisions.
What should your team establish first?#
- Scope: which information and supported tools the workflow needs.
- Ownership: who maintains the instructions and reviews the result.
- Approval: which actions need a person’s decision before execution.
- Provider policy: which account, training controls and retention terms apply.
Crevox is in prelaunch. If your evaluation requires specific access controls, retention commitments, a data-processing agreement or compliance evidence, contact contact@crevox.io to discuss those requirements. This page does not establish that those requirements are already met.