Prismian FAQs

Question 1

What is Prismian?

Accepted Answer

Prismian is an open-source access layer that establishes secure, permissioned, and audited access for AI agents to interact with an organization's internal data sources like Postgres, utilizing the Model Context Protocol (MCP).

Question 2

How does Prismian ensure data security for AI agents?

Accepted Answer

Prismian provides granular, agent-specific permissions, column-level redaction, and read rate limiting. It also maintains comprehensive audit logging for all agent data interactions. Connected collections are strictly read-only, preventing accidental writes to source databases.

Question 3

What data sources does Prismian currently support?

Accepted Answer

Prismian currently integrates with Postgres databases (including Supabase, Neon, and RDS). Support for Google Sheets and Notion is planned next, with Linear, Airtable, MySQL, and BigQuery also on the roadmap.

Question 4

Can AI agents write data back to my original database through Prismian?

Accepted Answer

No, AI agents cannot write to 'connected collections,' which are read-only mirrors of your source database. However, agents can create and update entries in 'native collections' within Prismian, designed for capturing AI-generated insights, notes, and decisions.

Question 5

What is the Model Context Protocol (MCP) and how does Prismian use it?

Accepted Answer

The Model Context Protocol (MCP) is a standard enabling AI tools (like Claude, Cursor, ChatGPT) to securely query and interact with data sources. Prismian is fully MCP-compatible, exposing a suite of secure tools (e.g., query_structured, search, write_entry) to these AI clients.

Prismian

Prismian

Key Features

Use Cases

Key Features

Use Cases