Security & privacy

Four privacy modes. You pick the tradeoff.

tableArth.ai is built for B2B software that ships to customers of every privacy posture — from cloud-native to fully local. Set the mode per customer, per workspace, or per data table.

Most capable

Full AI

Rich answers with chart and insight. LLM sees the full table.

Capability—
Privacy

Masked data

Text values tokenized before the LLM. Original values restored in output.

CapabilityPrivacy
Hybrid

Stats only

AI sees column statistics. Full table rendered locally — answer context uses statistics.

CapabilityPrivacy
Most private

Local template

Pure server-side rendering. Local answer formatting; query planning may still use a model.

—Privacy

These modes govern answer composition; query planning may still use an AI model.

Plus API key + origin guard — every widget locked to your domain.

Answer modes

Flexible for every use case.

Same engine, four operating profiles. Configure per customer, per table, or per plan tier.

Full AI

Complete AI answer with formatted markdown, bullet insights, data summary, and auto-selected chart.

LLM + full data

Masked data

Text values tokenized before the LLM. AI answers with masked data; original values restored in the output your users see.

LLM + tokens only

Stats only (Hybrid)

AI generates narrative from column statistics. Full data table rendered locally — rows never exposed to the model.

LLM + aggregates

Local template

Pure server-side rendering with deterministic templates. No answer-composition model call; query planning may still use a model. Maximum data privacy and infrastructure control.

Local answer rendering

Origin guard

Every API key is bound to your domains. Requests from any other origin are rejected at the edge.

Locked to your domain

Audit & access

Review workspace access, agent configuration, and available audit records with your administrator. Confirm identity-provider requirements for your deployment.

Workspace access · Agent settings
In practice

Set the policy. Inherit it everywhere.

You configure the mode at workspace level. The widget, the API, and the Chrome extension all inherit the same policy — your customers never see a less-strict surface by accident.

  • → Per-customer overrides for regulated tenants
  • → Per-table overrides for PII-heavy datasets
  • → BYO LLM key on enterprise — route through your own provider
  • → Security documentation available on request
Illustrative policy goalsworkspace · acme
defaults:
  mode: full_ai
  log_queries: true
  byo_llm: false

overrides:
  # EU enterprise tenants stay masked.
  - match:    customer.region == "EU"
    mode:     masked

  # Choose local answer formatting for PII.
  - match:    table.tags.contains("pii")
    mode:     local_template
FAQ

Privacy & security questions.

What are the four privacy modes?

Full AI uses query results to compose answers; Masked data tokenizes text values; Hybrid uses column statistics for the answer; Local template formats answers server-side. These modes govern the answer stage. Query planning can still use an AI model.

Can I set a different privacy mode per customer or table?

Configure the answer privacy mode for your agent and review the available controls with your administrator. Confirm the settings used by each integration before rollout.

Do you train on our data?

Data handling depends on your model provider and deployment configuration. Review the applicable retention and training terms with the team. Local answer formatting does not guarantee that no information leaves the query pipeline.

How do you stop a widget being used from another site?

Configure allowed origins and protect API credentials. Origin validation complements authentication; it does not replace server-side identity and access controls.

Can we route AI calls through our own LLM provider?

Model and provider options depend on your deployment. Contact us to review supported providers, keys, access controls, and rollout requirements.

Are you SOC 2 compliant?

Request current security documentation from our team. For the current status and to review our controls, book a security review and we'll walk through the specifics for your rollout.

Talk to us

Want to see the modes live?

We’ll walk through the exact policy you’d set for your customer base.

Book a security review See pricing