Legal · Subprocessors

Subprocessors

The third parties the NSTACK platform uses, what each one does, the data it handles, and where it applies.

Last updated
Derived from
The platform's code and configuration on that date

Where data goes

Platform · Where data goesDrawn from the platform's code. Your own systems connect to your instance under your agreements with them.

02·The list

Nine third parties, by name and purpose.

Seven are verified in the code. Two carry the to confirm label until their use in each instance is settled.

Subprocessors of the NSTACK platform
SubprocessorPurposeData involvedWhere it applies
RailwayHosting Runs every part of an instance, with its Postgres databases, its event stream, and its storage. Everything an instance holds: settings, connection settings, run history, the change log, uploaded files, and knowledge content. All instances
ClerkIdentity Sign-in, sessions, organizations and roles, member invitations, and machine tokens. One sign-in application for each front end and environment. Names, email addresses, organization memberships and roles, and sessions. Invitation emails go out from Clerk. All instances
AnthropicModel provider Serves agent runs on the platform model in production, and on your firm's own key. Drives Scan replays and reads recorded frames when your instance is set up with it. For agent runs: the agent's instructions and skills, the user's messages, knowledge search hits, and the rows the run's read tools return. For Scan: the pages a routine visits and the frames of a recorded workflow. Agent runs on its models · Scan
GoogleGemini APIModel provider Serves agent runs on the platform model in test and staging, and on your firm's own key. Scan uses it when it is the provider your instance is set up with. The same as Anthropic, for the runs it serves. Agent runs on its models · Scan when set up
OpenAIEmbeddingsModel provider Embeds knowledge fields, proposes queries and mappings in assisted mapping, and serves agent runs on your firm's own key. Scan uses it when your instance has no Anthropic key. The fields you choose to index for each record. For assisted mapping, a source's catalog and your business model. For agent runs, the same as Anthropic. Knowledge · assisted mapping · agent runs on its models · Scan when set up
PineconeVector search Stores and searches knowledge vectors, including the index assisted mapping searches. Embeddings of the indexed fields, a text preview of each, and the organization, environment, query, and record key as labels. Each organization and environment has its own separate space in the index. Knowledge · assisted mapping
BrowserbaseBrowser sessions Runs a cloud browser for Scan routine replays when your instance is set up for it. Otherwise routines run in a local browser. The pages a routine visits and the values it types into them. Scan
Amazon Web ServicesStorageTo confirm Storage in instances that have not yet moved to Railway storage, and the test environment's gateway until it moves. In those instances: connection settings, uploaded files, and knowledge content. To confirm, per instance
ResendEmailTo confirm Email delivery for app features that send mail, such as meeting invitations. Studio and Scan send none through it. Recipient addresses and the message. To confirm
Your systems

Warehouses, CRMs, and the sites a routine visits connect to your instance under your agreements with them.

Your own keys

When your firm adds its own model provider key, runs on that provider go through your firm's account.

Per instance

Separate vendor accounts and keys for each dedicated instance are on the security roadmap.

SourceThe libraries, settings, and configuration of every part of the platform, read on 2026-10-02

Ask for this list for your vendor review.

Write to us for the list as your instance is configured, and for the security review that goes with it.