Technical & Security overview

The assistant asks. Squark computes


How the Squark MCP Server authenticates, isolates, supervises, and audits every call — so your donor data stays yours.

ShareLinkedInXEmail

Authentication


AWS Cognito 'Sign in with Squark' — your team uses their own Squark accounts, never shared API keys. Enterprise sign-in aligns with the identity provider your organization already runs.

Isolation


Credentials scope per user and per organization. Tokens are encrypted at rest and rotated automatically. Every call is attributable to a specific user in a specific org — no shared service accounts.

Data boundaries


Donor data never trains foundation models. The assistant asks; Squark computes. Payloads returned to the conversation are the model's own predictions and explanations — not raw records exported to a third-party model provider.

Supervision


Human-in-the-loop by default. The assistant proposes; your team approves every consequential step — data registration, model configuration, scoring, and delivery. An expanded autonomous tool set exists for teams that opt in.

Tooling


17 curated tools spanning the full pipeline: register data, choose the outcome to predict, build the model, score a separate file, retrieve ranked results, explain the drivers behind any individual score. Each tool has a defined contract; the assistant cannot invent one.

ShareLinkedInXEmail

For IT & security

Questions your IT team will ask


Is our donor data safe?

You sign in with your own Squark account (AWS Cognito). Credentials isolate per user and per organization, encrypted at rest, tokens rotating automatically. Donor data never trains foundation models — the assistant asks; Squark computes.

Is this a chatbot on top of dashboards?

The opposite. The real engine is exposed as tools the assistant reads, writes, and executes. It never invents a score; every number comes from a model built on your data.

What can it actually do?

The full pipeline: register any dataset, choose the outcome to predict, build the model automatically, score a separate file, retrieve ranked results, explain the drivers behind any individual score.

What does the AI decide on its own?

Nothing consequential. Launch configuration is supervised: the assistant proposes, you approve. (An expanded autonomous tool set exists for teams that want it.)

Where does authentication live?

AWS Cognito, under your organization's Squark tenancy.

Can we scope which tools are exposed?

Yes — launch configuration is a supervised, curated tool set; scoping is part of onboarding.