Polytomic Harbor

Governed data environments for your AI agents

Let your AI agents securely query and act on data from your business applications, databases, data warehouses, spreadsheets, and cloud storage.

Book demo
SalesforcePostgreSQLSnowflakeAgent

Safe access and actions

Safely grant AI agents access to data across your systems. Each Harbor is isolated, with built-in access controls for humans and AI agents.

Fast to provision

Fast to provision through Polytomic’s no-code UI or MCP server. No data teams or engineering knowledge required.

Minimal token spend

Minimize LLM token spend. Agents query one fast store instead of hitting systems’ rate limits and constantly re-retrieving the same data.

Fully API-driven

Can be fully driven by Polytomic’s API, making Harbors perfect for enabling your AI agents to access customers’ systems.

Sync in the data you need

Polytomic’s sync engine fills the Harbor from SaaS applications, databases, data warehouses, spreadsheets, and cloud storage buckets. Sync filters restrict what comes in to what is relevant to the Harbor’s audience, and the same engine can write results back out.

  • Any Polytomic integration, in or out
  • Filtered at the source
  • Governed write-back to other systems
SalesforceSnowflakeZendeskPostgres

Connect your agents over MCP

Each Harbor comes with its own MCP server, so Claude, ChatGPT, Cursor, and other agents can securely query and act on its data. A per-Harbor access control list covers humans and agents alike: if you are not on it, your agent cannot connect.

  • One MCP server per Harbor
  • Access control down to the person or agent
  • Physically isolated from every other Harbor
ClaudeChatGPTYour agent

Give agents full context

Write the Harbor’s context document in plain prose: definitions, quirks, caveats, skills. Every agent that connects is automatically aware of it, so it starts querying and acting immediately, and accurately.

  • Plain-language definitions and caveats
  • One document per Harbor, team, or customer
  • Picked up automatically on connect
contextARRactive subscriptions onlyRegionbilling address, not HQChurnedstatus = closed_lostTicketsexclude internal

What’s in a Harbor

Data storage

A data storage container that scales to large data and is fast to query.

Syncs

Syncs going to or from the Harbor and any of Polytomic’s supported integrations.

MCP server

An MCP server with an access control layer dictating which people or agents are allowed to access the Harbor.

Context documents

Attached context documents instruct agents on how to accurately query the Harbor’s data.

Example scenarios from Polytomic customers

Revenue operations and GTM engineering

Sales reps would like to use AI agents (Claude, ChatGPT, etc) to query data about their accounts. This data may span Salesforce, Gong call transcripts, a Google Sheet, and a product table from Snowflake.

How RevOps sets it up:

  1. Create a Harbor per sales rep, region, or team.
  2. Activate a sync from those systems into each Harbor, with sync filters that restrict the data to what is relevant to that Harbor’s audience.
  3. Populate the Harbor context document with prose about your data’s quirks, caveats, and definitions.
  4. Specify a list of users or roles who can access this Harbor.
  5. Share the Harbor’s MCP server with the intended audience. Every connected agent is automatically aware of the context document, so it can start querying and acting immediately.

None of this involves the data or engineering teams. Because Harbors are physically isolated from one another, connecting agents cannot see anything beyond what you chose to sync in, which answers the security concern data teams usually raise about pointing AI agents at a whole warehouse.

Frequently asked questions

Give your AI agents a Harbor

We will provision your first Harbor with you and connect an agent on the call.

Book demo