# Security and Data Handling | Silicon

URL: https://meetsilicon.com/security/

> How Silicon handles your data: read-only warehouse access, queries that run in your warehouse, AI that inherits user permissions and no training on your data.

Security
# Your data stays in your warehouse
Silicon queries your warehouse with read-only access, applies every permission to every answer, and never uses your data to train AI models. This page describes how Silicon handles data today.
## Data stays in your warehouse
Silicon connects with read-only credentials and runs queries in your warehouse. Results are cached briefly to keep dashboards fast.
## Your data never trains models
Customer data is never used to train AI models. Answers are built from your semantic layer, not memorised.
## AI inherits your permissions
The AI analyst runs every query as the person asking, so row- and column-level rules apply to every answer.
## Deploy your way
Hosted in the US, with an EU region on Pro. Enterprise can run Silicon privately in your own VPC.
## How Silicon accesses data
Silicon is a query layer. It does not copy your warehouse.
Your warehouse Data stays here
Silicon Read-only queries, permissions applied
Your team Sees only what they're allowed
Read-only connections
Connect with a warehouse user or service account that only has read access to the schemas Silicon needs.
Queries run in your warehouse
Every chart, dashboard and AI answer becomes SQL that runs in your warehouse, under the credentials you provide.
SSH tunnels
Connect to warehouses in private networks through an SSH tunnel.
What Silicon stores
Metadata such as your project structure, metric definitions, and chart and dashboard configuration. Query results are cached for up to a day to keep dashboards fast, and query history is kept for 32 days.
## AI and your data
The AI analyst works through your semantic layer and inside the same permissions as everything else.
No training on your data
Customer data is never used to train AI models.
Model providers
Hosted Silicon uses OpenAI and Anthropic models. Self-hosted deployments can also use Azure OpenAI or Amazon Bedrock, keeping requests inside your cloud.
Answers through the semantic layer
The AI analyst queries governed metrics and models, not arbitrary tables. Raw SQL is only available to people whose role allows the SQL runner.
Permissions are inherited
Every AI query runs as the person asking, so their row- and column-level access applies. The AI never widens what someone can see.
Scoped agents
Admins can limit which models an AI agent can use, or run it in metadata-only mode so it answers from definitions without querying data.
AI retention
Conversation threads are kept so people can return to them; Enterprise can set a retention period. Logs of AI tool calls made through MCP are kept for 90 days.
## Access controls
Roles and permissions
Control who can view, explore, build and administer. Pro adds user groups; Enterprise adds a custom role builder.
Row- and column-level security
Restrict rows and columns using user attributes, applied to dashboards, exports, embeds and AI answers alike.
Private spaces
Keep work-in-progress dashboards and charts visible only to the people who need them.
SSO, SAML and SCIM 2.0
Single sign-on and automated user provisioning on Enterprise.
Support access
Silicon staff cannot see inside your workspace by default. Support access needs separate consent from your team, lasts 15 minutes per session, and every session is logged.
## Deployment options
Plan | Where it runs
Team | Hosted by Silicon in the US
Pro | Hosted by Silicon in the US or EU
Enterprise | Hosted, or private deployment in your own cloud or VPC
## Compliance status
Last updated 4 October 2026.
SOC 2
Not yet audited. We will publish the report status here once an audit is under way.
HIPAA
Not yet available. We do not currently sign Business Associate Agreements.
GDPR
An EU hosting region is available on Pro, and we sign Data Protection Agreements on request.
Security questionnaires
We're happy to complete your vendor security questionnaire. Email help@meetsilicon.com.
## Report a vulnerability
If you believe you've found a security issue in Silicon, email help@meetsilicon.com with "Security report" in the subject line and enough detail for us to reproduce it. Please give us a reasonable chance to fix it before sharing it publicly. We'll acknowledge your report and keep you updated as we work on it.
FAQ
## Security questions
### Where is my data stored? +
Your data stays in your warehouse. Silicon stores metadata such as your project structure and chart definitions, caches query results for up to a day to keep dashboards fast, and keeps query history for 32 days. Hosted workspaces run in the US, with an EU region on Pro.
### Can we self-host Silicon? +
Yes, on Enterprise. Silicon can run in your own cloud or VPC, so data and AI requests stay in your infrastructure, and product telemetry can be turned off.
### Which AI providers does Silicon use? +
Hosted Silicon uses OpenAI and Anthropic models. Self-hosted deployments can also use Azure OpenAI or Amazon Bedrock. Customer data is never used to train models.
### Is Silicon SOC 2 or HIPAA compliant? +
Not yet. We have not completed a SOC 2 or HIPAA audit, and we won't claim either until we have. We'll publish the status here as it changes.
### Do you sign DPAs and complete security questionnaires? +
Yes. Email help@meetsilicon.com and we'll send our Data Protection Agreement or complete your questionnaire.
## Questions from your security team?
Talk it through with the founder, or email help@meetsilicon.com for our DPA and questionnaire answers.
Talk to Sujay
30-minute call with the founder