Sub-processors

The third parties that process customer data for the NashOS application, what each one does, and the data it receives.

Last updated: 25 October 2026

Sub-processor	What it does for NashOS	Customer data it receives	When
Railway	Hosts the backend API and the database of each customer environment (a separate project per environment), and the onboarding wizard's backend and database.	All customer data at rest, and each environment's configuration secrets.	Always
Vercel	Hosts the web frontends (the application, the onboarding wizard and this website) and forwards the application's API traffic to the backend.	Customer data in transit between the browser and the backend.	Always
Anthropic	Provides the language model behind Nash, the in-app AI assistant, behind the summaries written for close runs, and behind the onboarding wizard's assistant and mapping suggestions.	From Nash: the user's messages and the conversation so far; the user's first and last name, username and role, and the Nash preferences they have saved; today's date and which years and months hold actual figures; and the results of the tools Nash calls, which carry the customer data it reads to answer (only what the asking user may see) and the draft cards it prepares, including the one-time password on a card that creates a user. Where conversation summaries are switched on for an environment, the older messages of a long saved conversation, with the summary so far, are sent to be summarised. From close runs, including scheduled ones that no user starts: when a run that carries the full close completes (every run a user starts, and a scheduled run set to run the full close), each step's result (row counts, totals, the largest variances and anomalies with their account names, any flagged figures with their account and entity, and the source's name); when any run is held, a scheduled sync included, the source's name, the hold policy, up to five flagged figures with their stored and loaded values, the statement totals before and after the load and the lines that moved most. From the onboarding wizard: the sheet name, header and up to four sample values of each column it cannot match; account codes, names, types and groups, and dimension member codes and names, while an upload is mapped; and, in wizard chat, each question with the workspace's name, industry, fiscal-year end and currency, its numbers of entities and accounts, its revenue, gross profit, EBITDA and net income for the latest year of actuals, and its most recent actuals month.	Whenever Nash or a close run is used, and when the wizard maps an upload or its assistant is used
Resend	Delivers the email the application and the onboarding wizard send, for example task assignments, review and approval notices, and close-run updates.	Recipients' email addresses and the content of each message.	Always
GitHub	Runs the automated jobs that provision a new customer environment, back up the app.nashos.ai database every night and test-restore a backup every month.	When these jobs copy a customer's onboarding data into a new environment's database, that data, including user accounts, passes through a GitHub-hosted runner on the way, together with the provisioning request (customer name, subdomain and owner email address). The nightly backup and the monthly restore test pass the app.nashos.ai database through a GitHub-hosted runner; the backup is encrypted on the runner before it is stored.	When an environment is provisioned; every night for the backup and every month for the restore test
Sentry	Records application errors so they can be fixed.	Error reports from the application: the error message and stack trace, the page or API address involved, browser and device details, the environment's name, and the signed-in user's internal ID and role. No name, username, email or IP address is stored; cookies, sign-in tokens and request bodies are removed before a report is sent.	When an error occurs in the app.nashos.ai environment
Cloudflare	Stores the nightly backups of the app.nashos.ai database (Cloudflare R2).	The database backup, encrypted before it is uploaded. Cloudflare does not receive the key.	Every night
Tavily Opt-in	Web search for Nash.	The search queries Nash writes, which can include parts of a user's question.	Opt-in. On for app.nashos.ai, NashOS's own workspace; on a customer environment only where the customer has opted in.

Railway

What it does: Hosts the backend API and the database of each customer environment (a separate project per environment), and the onboarding wizard's backend and database.
Customer data it receives: All customer data at rest, and each environment's configuration secrets.
When: Always

Vercel

What it does: Hosts the web frontends (the application, the onboarding wizard and this website) and forwards the application's API traffic to the backend.
Customer data it receives: Customer data in transit between the browser and the backend.
When: Always

Anthropic

What it does

Provides the language model behind Nash, the in-app AI assistant, behind the summaries written for close runs, and behind the onboarding wizard's assistant and mapping suggestions.

Customer data it receives

From Nash: the user's messages and the conversation so far; the user's first and last name, username and role, and the Nash preferences they have saved; today's date and which years and months hold actual figures; and the results of the tools Nash calls, which carry the customer data it reads to answer (only what the asking user may see) and the draft cards it prepares, including the one-time password on a card that creates a user. Where conversation summaries are switched on for an environment, the older messages of a long saved conversation, with the summary so far, are sent to be summarised.

From close runs, including scheduled ones that no user starts: when a run that carries the full close completes (every run a user starts, and a scheduled run set to run the full close), each step's result (row counts, totals, the largest variances and anomalies with their account names, any flagged figures with their account and entity, and the source's name); when any run is held, a scheduled sync included, the source's name, the hold policy, up to five flagged figures with their stored and loaded values, the statement totals before and after the load and the lines that moved most.

From the onboarding wizard: the sheet name, header and up to four sample values of each column it cannot match; account codes, names, types and groups, and dimension member codes and names, while an upload is mapped; and, in wizard chat, each question with the workspace's name, industry, fiscal-year end and currency, its numbers of entities and accounts, its revenue, gross profit, EBITDA and net income for the latest year of actuals, and its most recent actuals month.

When

Whenever Nash or a close run is used, and when the wizard maps an upload or its assistant is used

Resend

What it does: Delivers the email the application and the onboarding wizard send, for example task assignments, review and approval notices, and close-run updates.
Customer data it receives: Recipients' email addresses and the content of each message.
When: Always

GitHub

What it does

Runs the automated jobs that provision a new customer environment, back up the app.nashos.ai database every night and test-restore a backup every month.

Customer data it receives

When these jobs copy a customer's onboarding data into a new environment's database, that data, including user accounts, passes through a GitHub-hosted runner on the way, together with the provisioning request (customer name, subdomain and owner email address).

The nightly backup and the monthly restore test pass the app.nashos.ai database through a GitHub-hosted runner; the backup is encrypted on the runner before it is stored.

When

When an environment is provisioned; every night for the backup and every month for the restore test

Sentry

What it does: Records application errors so they can be fixed.
Customer data it receives: Error reports from the application: the error message and stack trace, the page or API address involved, browser and device details, the environment's name, and the signed-in user's internal ID and role. No name, username, email or IP address is stored; cookies, sign-in tokens and request bodies are removed before a report is sent.
When: When an error occurs in the app.nashos.ai environment

Cloudflare

What it does: Stores the nightly backups of the app.nashos.ai database (Cloudflare R2).
Customer data it receives: The database backup, encrypted before it is uploaded. Cloudflare does not receive the key.
When: Every night

Tavily Opt-in

What it does: Web search for Nash.
Customer data it receives: The search queries Nash writes, which can include parts of a user's question.
When: Opt-in. On for app.nashos.ai, NashOS's own workspace; on a customer environment only where the customer has opted in.

Data locations

Where each sub-processor stores or processes data is available on request: email security@nashos.ai.

Services that are not sub-processors

Google Workspace. Our company email, including messages you send to us. The application sends it no customer data.
Have I Been Pwned. Checks a new password against known breaches. It receives only the first five characters of the password's SHA-1 hash, with no account details.
Google Fonts. The application's web frontend loads some of its fonts from Google Fonts, which receives the viewer's IP address and browser details but no customer data.
The systems you connect. Your ERP, accounting, payment or file-transfer systems are your own data sources, under your own agreements with their providers.

What this list covers

This list covers the NashOS application and the data customers put into it. It does not cover this marketing website; for that, see our privacy policy. How the application itself is secured is on our security page.

Changes to this list

We update this page when the list changes, and the date at the top shows the latest change. For questions about this list, or about being told of changes in advance, email security@nashos.ai.