Use AI on your own data without sending it out.
Your apps and bots call one address. The answers come from models that run in house, not from a cloud AI company.
Who it is for, and the problem
Teams that handle other people’s data and still want AI: agencies, law firms, accountants and in-house teams that run several AI-driven apps.
- 01Using AI on client data usually means sending it to an outside AI company.
- 02Every app has its own AI key, so nobody can say who used what, or switch one off safely.
- 03One client’s data must never show up in work for another.
How it works
Every call passes the same few checks in the same order. Only the door is open to your apps; the model is never reached directly. In version 1 only local models answer.
Main features
All features →Any app, script or bot that speaks the standard chat API format most AI apps already use can switch to it by changing the address and the key. Answers in a fixed data shape are supported. The model is chosen by setting, so apps need no code change.
A person makes a key for each app, script or bot and sees it once. Keys can be suspended, revoked or handed to a colleague. There are no shared keys, so every call traces to one person.
Answers come from models running in house, on the server set up for you. No cloud AI provider is configured in version 1, so no prompt or answer is sent to one.
One client’s requests, records and cost never reach another client. Every record is fenced by person and by place (organisation, client). Our staff manage accounts and seats but never read prompts, answers or logged text.
By default only the facts of a call are kept: who, which model, how many tokens, when. Prompts and answers are logged only if your organisation and that app’s key both switch it on, and then only the key’s owner sees them.
Calls and tokens are counted per app key, per person and per organisation and shown on a usage page. Daily totals are meant to go to your IT Labs India account for billing; that link is still being connected.
Built and in testing. Anything that does not pass testing will be removed.
Security and where your data lives
- 01In version 1 every answer comes from a model on the server set up for you. No text is sent to a cloud AI provider.
- 02Prompts and answers are not stored unless your organisation and that app’s key both switch the log on. By default only who, which model, how many tokens and when are kept.
- 03Each app has its own key, shown once. Keys can be suspended or revoked without touching other apps.
- 04Records are fenced by person and by client in the database itself. Our staff manage accounts and seats but never read your prompts or answers.
- 05People sign in with a passkey, not a password.
NectarBox is in testing. It has not been independently audited, and we claim no compliance certification.
Works with
BeeBoxAI agents can use it as their model door (planned) The AI Hivecan use it for its own model work (planned) eCRMmail summaries and reply drafts (planned)Any other app that speaks the standard chat API format can call it too.
What you need to run it
A server that can run small open AI models. We decide the size with you: it depends on which models you want and how fast answers must be. On our test box, small models run on an ordinary server processor, and an answer takes from a few seconds to about a minute.
A server with a dedicated graphics processor would give faster answers and room for larger models.
NectarBox runs in IT Labs India’s cloud or on your premises, as a private AI box. Either way it runs on its own, with its own database and sign-in. Version 1 runs on our test box today.
How a pilot works
Cost is scoped per pilot. Talk to us.- 01Talk to us. Tell us which apps, bots or scripts should use AI, and what data they handle.
- 02We set up the box. We set up NectarBox and the local models with you, with your organisation.
- 03People join. Each person gets a seat and a welcome link, and makes a passkey instead of a password.
- 04Apps connect. People make a key for each app or bot and change its address and key. The app is otherwise unchanged.
- 05We review together. You see usage by person and app, and decide what to keep.
Ready today
Built and in testing
Nothing is marked “Available now” yet. Version 1 runs on our test box. Built and in testing. Anything that does not pass testing will be removed.
- In testingStandard chat door
- In testingA key for every app or bot
- In testingLocal models only
- In testingClient walls
- In testingText log only when switched on
- In testingUsage per person, app and organisation
- In testingPasskey sign-in and welcome link
Try NectarBox with your team
A pilot, set up with you and scoped to your apps.