Connects AI to your systems, and nothing more.

Interstellar sits between your assistant and the file shares, databases and applications your teams already use. It brings back what a person is allowed to see, and nothing else.

Start a pilotHow it works
File shareSharePointPostgreSQLS3 archiveSQL ServerPermission checkFinance analyst3 of 5 sourcesHR officer3 of 5 sources
Illustration. The same question reaches five systems; each person only receives what they could already open.

Permissions come with the data.

Access rules from each source system travel with everything Interstellar indexes, and are checked on every request, before anything is ranked or shown.

  • Same question, different answers

    A finance analyst and an HR officer can ask the same thing and get answers built only from what each of them may read.

  • Acts as the person where it can

    Where a system supports it, Interstellar signs in as the person asking, so the system itself enforces their access.

  • No hints from hidden files

    Results the person can’t open are removed before ranking, so even the count of matches never leaks.

Nothing changes without a yes.

Reading is automatic. Anything that would change a record, send a message or start a process stops and waits for the person to approve exactly what will run.
  1. 01

    Proposed

    The assistant suggests an action, such as raising a ticket, with its exact details.

  2. 02

    Shown

    The person sees the system, the action and every value it will use.

  3. 03

    Approved

    Their approval covers that exact action. Change one value and it has to be approved again.

  4. 04

    Run

    Interstellar carries it out with the person’s own permissions.

  5. 05

    Recorded

    Who, what, when and the outcome go into the audit log.

Connects to what you already run.

Each connector declares what it can list, search and fetch, and which of its actions only read and which make changes. Admins switch each one on or off by group.
File shares
Local and network shares, with their folder permissions
Object storage
S3-compatible stores such as MinIO
Databases
PostgreSQL and SQL Server, read-only, limited to schemas and views an admin allows
Business applications
Any REST API described by an OpenAPI specification
Microsoft 365
SharePoint and OneDrive, where your boundary allows it
MCP servers
Existing Model Context Protocol servers plug in as connectors

Treats everything it fetches as untrusted.

A document can inform an answer. It can never give instructions, so text hidden inside a file can’t make the assistant act.

  • Sealed credentials

    Connection secrets are encrypted at rest with a key you hold, and never reach the browser, the model or the logs.

  • Every call on record

    Each search, fetch and action is written to an append-only, tamper-evident log you can stream to your SIEM.

  • One layer, many products

    AE-1 and future Sharq products use the same connectors, rules and audit trail.

Questions integration teams ask.

Do we have to copy our data somewhere new?

No. Connectors can query systems live, or index content for faster search; either way it stays inside your infrastructure.

Can the assistant change our systems on its own?

No. Every action that changes something needs a person to approve it, and admins can disable any action for any group.

How are permissions kept in sync?

Indexed content stores the source system’s access rules and they are checked at query time. Where possible, Interstellar acts with the person’s own credentials instead.

Works with

AE-1A private AI workspace for teams that work in Arabic and English.Sharq EngineTrains and serves models on your data, on your infrastructure.All productsHow the three work together, on one page.

Start with one workflow.

A pilot runs on your own data, inside your boundary, and is measured against the numbers you already report.

Start a pilotRead the client briefing