Security for a company AI workspace.
Kapella keeps work inside the company. Model calls run on the server, chats stay with your company’s members, and unsafe requests are blocked. You should still review generated output before you rely on it.
How Kapella protects company work
Built for paid companies that will not hire an in-house AI team, not for anonymous public chat.
Company isolation
Only active members of your company can read that company’s chats, playbook, members, and usage. Other companies cannot see your workspace.
Keys stay off the browser
Paid model calls run on the server. Browser apps never hold provider keys. Your team does not manage API credentials.
Safety before and after
Requests are screened before generation and answers are screened after. Jailbreaks and clearly disallowed content are blocked and logged.
Prompts stay with the company
Full prompt and response text lives on the company chat record. Operational logs keep metadata, not the full conversation. We do not use customer prompts to market Kapella.
No unsafe fallback
If a request is blocked for safety, Kapella does not retry it on another model. The refusal is the end of that request.
Admins control access
Admins invite seats, manage billing, and see remaining credits. Platform operators can inspect aggregated usage, not other companies’ chat transcripts.
What this means day to day
Sign in with Google or email and password. Admins invite seats. Members ask work questions. Private chats stay with the owner and admins. Single sign-on is not in this version.
Write only what you are allowed to send
Prompts go to third-party model providers on your behalf. Do not paste secrets or personal data you are not allowed to share with those providers.
Review before you send work onward
Kapella helps draft and summarise. A person in the company should still check the output, and can ask a teammate to review a thread.
A credit cap the admin can raise
Usage is metered. Admins see remaining credits and can raise the cap by upgrading. If it is reached, new prompts pause until then or until the month resets. Existing chats stay.
Security questions
Read the privacy notice for retention. Book an audit if you want help spotting which workflows belong in the workspace.
Full prompt and response text stays on the company chat record. Operational logs keep metadata, not the full conversation, unless a support investigation requires otherwise. We do not use customer prompts to market the product.
No. Paid model calls run on the server. Browser apps never hold provider keys. The workspace chooses a suitable model; your team does not manage API keys.
Safety screening runs before and after generation. Jailbreaks and clearly disallowed content are blocked and logged. There is no fallback to another model after a safety block.
People sign in with Google or with email and password. Single sign-on is not in this version.
Start inside a company workspace.
Create a paid company account, or ask about an AI workflow audit if you want help choosing the first jobs to put in Kapella.