Private AI for regulated enterprises

Your documents. Your model.
Zero data retention.

ironindex answers questions from your internal document vault with verifiable citations — while enforcing per-document access control and never logging a single prompt.

Zero-data-retention Per-document RBAC Self-hosted inference Inline citations

The platform

Built for information barriers, not just search.

Law firms and financial institutions can’t bolt AI onto a data set with ethical walls and need-to-know access. ironindex enforces those controls at the retrieval layer — where it actually matters.

Hard access control

Every chunk carries an access list. Retrieval is filtered inside the vector store, so a user can never receive — or even rank — content their role doesn’t permit.

Zero data retention

Prompts and completions are never written to disk or telemetry. The gateway redacts message bodies and no chat history is persisted server-side.

Cited, grounded answers

Answers are built strictly from retrieved context, with inline source and page citations you can click to preview the exact passage.

Bring your own model

Swap between self-hosted GPUs and cloud endpoints with a one-line change. Inference stays inside your perimeter — your data never leaves.

Access control

Access control that survives the ethical wall.

Permissions aren’t a prompt instruction the model might ignore. Every query is bound to the user’s groups and enforced as a filter in the vector database — so unauthorized content is excluded before it can influence the answer.

# enforced on every retrieval, per user
filter = Filter(must=[
  FieldCondition(
    key="permitted_groups",
    match=MatchAny(any=user.groups),
  )
])
# no match → the chunk is never returned

Your documents. Your model.Zero data retention.