How it works | AlienGiraffe
How it works

Request, approve, sandbox, query.

A person asks for exactly the tables, columns, rows, or time window they need. An approver signs off. A short-lived sandbox next to the source loads just that slice and answers the queries. Raw rows stay inside the perimeter.

The architecture

Everything runs on your side of the line.

The AlienGiraffe Connector (requests, approvals, audit) and every sandbox run on infrastructure you own. The execution point sits next to the source, not in our cloud and not in a data platform's.

Your network

Sources, untouched

Oracle core banking
Postgres, MySQL, SQL Server
Log files on SSH servers
Snowflake, S3
read once

Sandbox, per request

Starts on approval, next to the source. Loads only the approved slice.

5 columns1 account14:00–15:00
queries

Approved consumers

People · Explorer
AI agents · MCP
Apps and vendors · API

AlienGiraffe Connector · your VM or Kubernetes

Request Approve by role Audit every step

Raw rows don't cross the line. Not to us, not to a cloud.

Crosses only if you send it: query logs to your S3 or Postgres.

Four steps

From question to answer without a ticket.

The same flow for a support engineer, an analyst, or a data scientist. Only the template changes.

  1. 01

    Request

    Pick the datasource, then the tables, columns, and row filters you need. Templates turn repeat requests into one click, and can ask for a date window.

    transactions · 5 columns · account_id = 4471 · yesterday

  2. 02

    Approve

    Approvers are set by role, by your admins. Low-risk requests can auto-approve. Each request shows a risk score from how many sources and fields it touches.

    risk: low · approver: support lead · rule: support-template

  3. 03

    Sandbox

    On approval, a short-lived sandbox starts inside your network, next to the source, and loads only the approved slice. The source is read once.

    sandbox up · 1 table · 5 columns · 212 rows

  4. 04

    Query

    People query in the explorer, in SQL or plain English. Apps, AI agents, and vendors connect through the API or MCP. Every query runs in the sandbox, not on production.

    “which transfers failed after 14:00?”

What a request names

A request describes a scope. Anything it doesn't name isn't loaded into the sandbox, and the approver sees exactly that scope before approving.

Datasource
Which system: core banking on Oracle, a Postgres replica, a log server.
Tables and columns
Only the columns the task needs. The rest never load.
Row filters
One account, one branch, one product line. Applied when the sandbox loads.
Date window
Templates can ask the requester for a window: yesterday, last full month, 14:00 to 15:00.

Policy engine

Your access rules, checked before anything loads.

Every request is evaluated against policy before a sandbox reads a single row. Start with policy as code, or keep policy where it already lives.

Default

Policy as code

Approval rules live in a config file you review and version with the rest of your infrastructure: who can request what, who approves it, and what auto-approves.

  • Rules by requester role
  • Roles mapped from your identity provider
  • Auto-approve for low-risk templates
Design partner program

Unity Catalog

Already keep access policy in Unity Catalog? Use those grants to decide who may ask for what, so data that never reaches Databricks follows the same rules.

Alien Giraffe with Databricks
Built with you

The policy engine you run

Immuta, OPA, groups in Okta or Entra, or rules your team wrote in-house. We connect Alien Giraffe to what you already run, starting from your existing policies.

Talk to an engineer

In the product today

Product features

Templates

Repeat requests become one click. A template fixes the source, tables, and columns, and can leave the date window to the requester.

Virtual tables

A template can join the selected tables into one virtual table, so the analyst queries one thing instead of rebuilding the join.

Risk score

Each request shows a score from how many sources and fields it touches, so approvers see the big ones coming.

Approval rules by role

Admins decide who approves what, by requester role. Low-risk templates can auto-approve.

Explorer, API, and MCP

People use the explorer: SQL with sorting, filtering, and paging, or plain English turned into SQL. Apps, AI agents, and vendors connect through the API or MCP.

Organizations and SSO

Multiple organizations in one deployment, with single sign-on for requesters and approvers.

For the DBA

What we ask of your database, and what we don't.

We need

  • A read-only login. On Oracle, a read-only account works.
  • A network path from the sandbox host to the source, inside your network.
  • One read per approved request, scoped to the columns and rows it names.

We don't need

  • × Supplemental logging, archive-log changes, or LogMiner grants.
  • × Change capture or replication on your primary.
  • × Write access of any kind.

Good to know

  • Your database audit trail sees the Alien Giraffe login. Alien Giraffe records who asked and who approved every request.
  • Load is bounded by design: one read per approved request, with follow-up queries in the sandbox.

Datasources

Databases and the logs next to them.

Most data access tools stop at SQL. Support and operations questions usually need a log file too.

Oracle

A read-only login works

PostgreSQL

MySQL

Microsoft SQL Server

Snowflake

S3

Log files over SSH

Including compressed files

Deployment

One VM, or your Kubernetes.

No SaaS dependency in the data path. Pick the shape your platform team already supports.

Single-VM bundle

Docker or Podman on one host. No registry access needed, which suits networks with no route out.

Kubernetes

A Helm chart for the cluster you already run.

Configuration as files

Datasources and request templates live in files, so they're reviewed and versioned like the rest of your infrastructure.