ironindex plugs into the tools your people already use and gives the model secure context from your own knowledge — read where it lives, with their credentials, answered by American open-weight models. None of it is copied to us, indexed by us, or used to train anything.
Most tools want your proprietary information moved into an index they control, then mirror your permissions on top and hope the two stay in step. ironindex connects to your systems where they already are, and holds none of it.
Document stores, mail, and your internal tools are reached where they already live — nothing is migrated, mirrored, or re-permissioned. Adding a system is a connection, not a data-transfer project.
Every call runs with the individual's own credentials, so each system decides what that person may see — on every request. We never mirror an access list, so nothing drifts out of step and nothing needs re-syncing when someone changes role. With Google Drive it is narrower still: only what you hand us through Google's own picker.
Beyond reading, the model can use your internal tools through the same connection — looking things up and, where you allow it, acting. Every tool is on an allowlist you control, and write access is off until an operator grants it.
Answers come from NVIDIA's open-weight models — developed by a US company and licensed for commercial use. We keep every customer on the strongest one we can serve, so nobody has to evaluate a model, negotiate with a lab, or migrate when something better arrives.
Retrieval runs in the browser, against your own systems, with the signed-in person’s delegated token. Nothing is copied, indexed, or routed through our infrastructure — which is why there is nothing for us to lose, and nothing to take our word for. Open the network tab and look. Every answer cites the passage it came from, and says so plainly when it found nothing rather than inventing one.