Security, privacy and data
How your data is handled.
Vera and Clara follow one policy. Plugin functions run inside Codex; Mparanza-hosted services form a separate processing boundary.
Vera + Clara · Video
How Vera and Clara handle data.
See what can stay on your computer, what may enter a language-model call through Codex, and when Mparanza becomes a separate processing boundary.
Watch on YouTube
When Vera and Clara work inside Codex.
Vera and Clara do not automatically anonymise data. They may use local Python to filter or aggregate information when useful. Data supplied to the model is processed through the user's existing ChatGPT plan. Workflows inside Codex do not send client files, prompts, or model-context content to Mparanza.
Workflows inside Codex send no client or work content to Mparanza.
Local processing is used when it helps the work.
Local Python can sort, calculate, reconcile, filter, aggregate, and create outputs without first moving complete source files to a separate Mparanza system.
This is not automatic anonymisation. When the professional task requires names, documents, original language, or case facts, that material may enter the model context.
Mapped once per workflow, not once per prompt.
Each workflow inside Codex is reviewed when it is added or changed. The review records what normally stays local and what Codex may read. It does not create a form, consent step, or record for each prompt.
Never put passwords, API keys, authentication cookies, access tokens, or session material in prompts or files Codex can read.
Gmail and WhatsApp Desktop use different boundaries.
Vera and Clara work in ChatGPT with material supplied in the conversation and with callable connected apps. After providing useful work, they may recommend Codex Desktop for direct folder access, persistent project files, local tools, and durable deliverables. Installation is optional and the conversation can continue in ChatGPT.
In ChatGPT or Codex, Vera searches Gmail through OpenAI's separately installed and connected Gmail connector. Mail stays in Gmail; Vera does not create a mailbox copy or store credentials or messages.
WhatsApp is inspected only from Codex Desktop with Computer Use in the WhatsApp Desktop app already opened and authenticated by the professional. No Mparanza server receives or stores a copy of those messages. Screen text and images read by Codex may still enter the model context under the user's ChatGPT/Codex account, and the terms of OpenAI, Gmail, and WhatsApp apply separately.
Mparanza-hosted services are a separate boundary.
When a Vera or Clara function uses a Mparanza-hosted service, the content needed for that service reaches Mparanza-controlled systems. Hosted interviews, Hosted Voice, and the retail-data bridge are examples, not separate policies.
Each hosted service is documented once at service level: what may be sent, who can access it, and its retention and deletion arrangements. There is no prompt-by-prompt documentation.
A public search, connector, portal, or send action chosen by the user follows that external service's terms. It is an external destination, not a third Mparanza processing category.
The plugins also contact Mparanza to check for updates and the status of previously submitted feedback. Those requests contain no client or work content, although technical connection records may still be logged. Feedback content is sent only through the explicit submission workflow.
One policy for Vera and Clara.
The distinction is architectural, not professional. A Vera reconciliation and a Clara presentation both fall in the first category when they run inside Codex.
Any Mparanza-hosted service used by either plugin falls in the second category and is covered by its service-level description.
Verify the position.
You do not have to rely on the claim alone.
One policy for Vera and Clara. No prompt-by-prompt paperwork.