Trust & security
How we handle your data.
Ataski processes your data only to deliver the service you signed up for. We do not sell it, we do not train AI models on it, and we do not share it with advertising networks or data brokers. Our public marketing pages use a Google Ads tag for conversion measurement only: no remarketing, no personalised ads (see Your Privacy Choices in the Privacy Policy). Below: where your data lives, who processes it, and what has and has not been independently audited.
Last updated 2026-10-07. The sub-processor list here changes the day it changes; a new sub-processor is announced to customers 30 days before it starts.
The facts on this page are read from the running system each time the page loads; last checked against the full repository on 2026-10-13.
At a glance
- AI training on your data
- Never
- Neither Ataski nor Anthropic trains models on it.
Default posture
We do not train AI on your data.
We use Anthropic's commercial API, which does not train on customer content. We have not opted in to training and will not. Anthropic may keep inputs and outputs for up to 30 days for abuse monitoring, under its commercial terms.
Default posture
Read by default. Every write is named.
We do not write to your books. Accounting systems (NetSuite, Sage Intacct, Xero, QuickBooks, Zoho Books, Microsoft Dynamics 365 Business Central and others) and Stripe are read-only in our code: Ataski issues read requests only and has no code path that posts a journal entry, an invoice or a payment. Where we do write, we name it: your mailbox sends a message you approved; Slack posts a digest or a direct message; Salesforce write-back logs activity on a contact; the Google Sheets and Excel add-ins write into the spreadsheet you opened them in. Salesforce write-back is off until you switch it on, never overwrites a field that already holds a value, and every write is recorded in the audit log. We do not request calendar write access: Google Calendar is free/busy only, and Microsoft 365 calendars are read-only.
SOC 2
Where the audit stands.
No SOC 2 audit has been performed yet: Ataski holds no SOC 2 attestation and has not engaged an auditor. The controls below are in place, and each one says what backs it.
This line changes when the record changes, and not before. We do not print a badge we have not earned.
Controls
What is in place, and what backs it.
Each claim below is checked against the running system when this page loads.
Where a claim rests on a provider's terms or on a written document rather than on running code, the label says so. Amber marks a control that is not in place yet: it is named here, and not counted.
35 of 37 controls in place.
Access
-
Your workspace cannot see another workspace's rows: the boundary is enforced by the database, not by application code.
What the code says: Every table that holds customer data (322 today) carries a Row Level Security policy, and the build fails if one is missing.
-
Sign-in is handled by WorkOS. We never see or store your password.
What the code says: The session cookie is host-only, Secure, HttpOnly and sealed, so a browser never sends it over plain HTTP or to another domain.
-
Who can reach your workspace is reviewed every quarter, and each review is signed off.
Scheduled: Quarterly access reviews are set up; the first one fell due on 2026-10-01 and has not been signed off yet, so this control is not counted.
-
We ask each system you connect for the narrowest grant it offers, and we name the ones that can write.
What the code says: 46 connections are listed below with the exact access each one requests: 22 can only read, and the 24 that can write say what they write.
-
Credentials are kept out of the code, and automated scanning stops an accidental leak before it reaches the repository.
What the code says: Secret scanning runs on every change before it is pushed and across the full repository history; credential settings are typed so their values never print in logs.
Data
-
Every action is written to an audit log that nothing can edit afterwards.
What the code says: The application's database role can add audit records and read them, but cannot change or delete them: the database itself refuses.
-
Credentials you connect are encrypted, used only to call the service they belong to, and never included in an export, not even your own.
What the code says: Credential tables are excluded from every data export, and encrypted credential columns are stripped from the rest.
-
Deletion is a purge with a reversible window, not a flag on a row.
What the code says: A 30-day grace period after you ask, then a hard purge of the workspace across every customer-data table and our file storage.
-
Your data leaves with you: one machine-readable export, on demand.
What the code says: One machine-readable export covers your workspace data; only stored credentials and data that is not yours (public-register copies, shared caches) are left out.
-
Leaving works: you take everything of yours first, then everything goes.
What the code says: Export at /api/export; after a deletion request, a 30-day grace period, then an automatic job deletes the workspace. Automated tests keep this path working.
-
Files you send us are checked before anything opens them.
What the code says: Every upload passes one intake check (file type and size) and a ClamAV virus scan; an infected file is refused and logged. If the scanner is briefly unavailable, the file is accepted, marked as unscanned and logged.
-
Company-level patterns we cache expire; contact records stay inside your workspace.
What the code says: The shared cache of company email formats (domains and address patterns, never individual contacts) is purged after 30 days.
Infrastructure
-
Your workspace data is stored in the United States.
Per the written document: Application: Hetzner Online GmbH. Database: Neon, Inc.. Files: Cloudflare, Inc.. A few vendors process limited data elsewhere; each one's location is in the sub-processor table.
-
Backups are tested by restoring them, because a backup nobody restores is not a backup.
Scheduled: The database is backed up continuously by our database provider (point-in-time recovery). The monthly restore test has been set up since 2026-09-06 but has not run yet, so this control is not counted.
-
Every page is served with locked-down browser security headers.
What the code says: 6 security headers on every response: Content-Security-Policy, Permissions-Policy, Referrer-Policy, Strict-Transport-Security, X-Content-Type-Options, X-Frame-Options. The Excel and PowerPoint add-in panes are the one exception to framing: Microsoft Office has to be able to display them.
Change management
-
Every change is checked for security issues before it ships.
What the code says: Before any change is pushed to the code repository, it passes 4 automated security checks: secret scanning of the change and of the full repository history, dependency vulnerability scanning and API fuzzing.
-
Production runs one reviewed version of the code, deployed whole.
What the code says: The deploy tool refuses to ship anything other than the reviewed main branch.
Monitoring
-
There is a written plan for the bad day, and a deadline inside it.
Per the written document: A personal-data breach is notified within 72 hours of us becoming aware; the plan was last updated 2026-10-08.
-
Errors and downtime reach us without a customer having to tell us.
What the code says: Bugsink receives application errors; Better Stack watches the public endpoints and the scheduled jobs.
Vendors
-
Every vendor that touches your data is named and bound by a written DPA.
Per provider terms: 16 sub-processors, list effective 2026-10-07. Its fingerprint (b6fcf738e3f0) is recorded on every signed DPA, so you can prove which version you agreed to.
-
Your text is processed only by Anthropic's Claude models, and Anthropic does not train on it.
Per provider terms: Models in use: Claude Haiku 4.5, Claude Sonnet 5, Claude Sonnet 5.5, Claude Opus 5, Claude Opus 5.5, Claude Fable 5.1. High-stakes output is reviewed by a separate, larger Claude model. No other provider's model drafts or reviews your text.
Written policies
-
The umbrella policy: what we protect, who answers for it, and the rules everything else hangs off.
Per the written document: Information Security Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written rule for who may reach production, how access is granted and how it is taken away.
Per the written document: Access Control Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written rule for how a change reaches production: one reviewed ref, whole, never a hand-copied file.
Per the written document: Change Management Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written standard for how the code is typed, tested and gated before it can reach you.
Per the written document: Secure Development Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written plan for the bad day: who commands, who is told, and by when.
Per the written document: Incident Response Policy, approved 2026-10-07; reviewed at least once a year and after any incident that contradicts it.
-
A written rule for adding a vendor, disclosing it to you and reviewing it.
Per the written document: Vendor Management Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written schedule for how long each kind of data is kept and when it is destroyed.
Per the written document: Data Retention Policy (internal), approved 2026-09-05; reviewed at least once a year and after any incident that contradicts it.
-
A written plan for coming back: backups, restore targets and how the service runs degraded.
Per the written document: Business Continuity and Backup Policy, approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written rule for how the operator may use the systems and the machine that reaches your data.
Per the written document: Acceptable Use Policy (internal), approved 2026-10-08; reviewed at least once a year and after any incident that contradicts it.
-
A written rule for which encryption we use, where the keys are kept and how they are replaced.
Per the written document: Cryptography and Key Management Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written rule for how security flaws are found and how fast each one must be fixed.
Per the written document: Vulnerability Management Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written method for listing what could go wrong, weighing it and deciding what to do about it.
Per the written document: Risk Management Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written rule that every server, service, account and device touching your data is listed and has an owner.
Per the written document: Asset Management Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written rule for destroying data when a disk, a device or a cloud resource is retired.
Per the written document: Media and Equipment Disposal Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written rule for how our controls are assessed and how every finding is tracked until it is closed.
Per the written document: Audit and Assurance Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
-
A written rule that your data leaves in open formats, over standard interfaces, whenever you ask.
Per the written document: Interoperability and Portability Policy, approved 2026-10-08; reviewed at least once a year and after any significant change.
Data location
Where your data physically is.
Three vendors hold your data at rest, all in US data centers. Each is a named sub-processor under its own DPA.
| What | Vendor and country |
|---|---|
| Application server | Hetzner Online GmbH · Germany (data center USA) |
| Database | Neon, Inc. · USA |
| Files and documents | Cloudflare, Inc. · USA |
Encryption
In transit and at rest.
- In transit: TLS on every connection (TLS 1.2 minimum, TLS 1.3 preferred) with certificates from a public certificate authority, and HTTP Strict Transport Security for two years across all subdomains, so after the first visit a browser will not use plain HTTP.
- At rest: the database (Neon) and file storage (Cloudflare R2) are encrypted with AES-256 by those providers, and the database's point-in-time backups are encrypted the same way.
- On top of that, credentials you connect are encrypted in the application, under a key held outside the database, and decrypted only at the moment of use.
Retention and deletion
How long we keep it, and how it goes.
Ask to delete the workspace and a 30-day grace period starts, during which you can change your mind. After it the purge is hard: every customer-data table and our file storage, the audit trail included. Copies that vendors keep on their own timers (AI request traces, error reports, email delivery logs, meeting recordings) expire under each vendor's retention terms, listed in the Data Retention Policy.
Before that, take everything: the export is machine-readable.
The shared cache of company email formats expires on its own timer and never holds an individual contact.
The full text is the Data Retention Policy. Data Retention Policy
Sub-processors
Vendors that process your data.
Customer data is sent only to the vendors listed here, each under a written data processing agreement. Each row says what the vendor receives and where it processes it.
| Vendor | Purpose | Data accessed | Region | DPA |
|---|---|---|---|---|
| Hetzner Online GmbH | Application server hosting | all customer data via application access | Germany (data center USA) | View → |
| Neon, Inc. | Postgres database hosting | all customer data via RLS | USA | View → |
| Cloudflare, Inc. | Authoritative DNS (records not proxied) and R2 object storage for files | uploaded documents (contracts, support attachments); generated PDFs (board packs, reports); HTML snapshots; meeting transcripts | USA | View → |
| WorkOS, Inc. | Customer sign-in and session management (email, passwordless, multi-factor) | authentication credentials; user identity | USA | View → |
| ActiveCampaign, LLC (Postmark) | Transactional email delivery and inbound email | recipient email addresses; email body content | USA | View → |
| Anthropic, PBC | AI model inference (Claude API) for drafting and review | user prompts; model outputs | USA | View → |
| Hyperdoc Inc. (d/b/a Recall.ai) | Meeting bot identity + transcription for any role that joins a live meeting on the customer's behalf | meeting audio; meeting transcripts; participant names and emails; meeting chat messages | USA | View → |
| Bright Data Ltd | Public-web search and company page extraction for prospect research | customer search criteria (industry, location, keywords); public-web search results; company and team page extractions | Israel (global data centers) | View → |
| Perplexity AI, Inc. | Public-web search and summaries used to personalise outreach | email or domain inputs; public-web summaries | USA | View → |
| ZeroBounce, Inc. | Email deliverability verification (pay-as-you-go) | email addresses submitted for verification, kept by ZeroBounce only as its DPA allows | USA | View → |
| Apify Technologies s.r.o. | Outbound prospect discovery from public LinkedIn data, run by Ataski under its own Apify account with no LinkedIn login: company search, the website a company publishes on its LinkedIn page, and people-at-company identity records (no email, no phone) | customer search criteria (keywords, industry, location, job titles); public company records, including LinkedIn company page URL and website; public professional identity records (name, job title, employer, LinkedIn profile URL, location) | Czech Republic (EU) | View → |
| Stripe, Inc. | Billing and payment processing (checkout, customer portal, sales tax) | billing details; subscription data | USA | View → |
| Langfuse GmbH | AI request tracing for quality monitoring (prompts and completions) | LLM prompt content; LLM completion content; model and cost metadata | Germany (data center USA) | View → |
| Bugsink B.V. | Error tracking (personal data scrubbed by default) | stack traces; request metadata (workspace and request identifiers) | Netherlands (EU) | View → |
| PostHog, Inc. | Product analytics (server-side sign-up and activation events) | event names and timings; workspace identifier (no names, emails or content) | USA | View → |
| Better Stack, Inc. | Uptime monitoring and heartbeats | endpoint health metadata | USA | View → |
Access controls
Connector scopes
Every operator-connectable system runs against a documented set of provider-side scopes. The list below renders from our internal scope manifest — what you see is what the OAuth client (or equivalent credential) actually requests.
-
Amplitude Reads only; the grant would allow writes
Reads, per customer, event counts, the date of last activity, and the most used events through Amplitude's Dashboard REST API, for the Renewal Manager's account-health signals, using your project's API key and secret key. Ataski issues read requests only — note that Amplitude offers no read-only project key: the same API key can send events into the project, so it is technically read and write.
-
Anthropic (Bring Your Own Key) Can write
Routes Claude calls through your own Anthropic API key. The key is stored encrypted and used only at the moment of a call; Anthropic bills you directly for that traffic, and Ataski's invoice leaves it out. Rotate or remove the key in Settings at any time.
-
BambooHR Reads only; the grant would allow writes
Reads your employees with job, pay rate, and pay period through a report run on demand (nothing is saved in BambooHR) so the Financial Analyst's headcount register can show staff and pay. Uses your company subdomain and an API key from a BambooHR user. Ataski issues read requests only — note that BambooHR offers no read-only API key: a key carries the permissions of the user who created it, so it is technically read and write wherever that user can write.
-
Bitrix24 Reads only; the grant would allow writes
Reads your deals (crm.deal.list — the only method called) so the Financial Analyst can plan revenue from the pipeline, through an inbound webhook you create in your own Bitrix24 portal with the CRM (crm) permission and can revoke in one click. Ataski issues read requests only — note that Bitrix24's crm permission has no read-only variant, so the webhook is technically read and write.
-
Chargebee Read only
Reads subscriptions and invoices so the Financial Analyst can build recurring revenue, with a read-only API key from your Chargebee site (Settings → Configure Chargebee → API Keys). Cannot create, change, or cancel a subscription or an invoice.
-
Close Reads only; the grant would allow writes
Reads opportunities for renewal dates, and searches leads so the Outbound Sales Assistant does not contact someone already in your CRM, with an API key you paste in. Ataski issues read requests only — note that Close offers no read-only API key: the key carries the full access of the user who created it, so the permission is technically read and write.
-
Deel Read only
Reads your people and each employment's pay rate and pay period so the Financial Analyst's headcount register can show staff and pay, with an API token you create in Deel (Developer → API keys). The token holds whatever access you give it in Deel; Ataski only reads the people list and never changes a contract or a payment.
-
DocuSign Read only
Reads completed envelopes and their signed documents for contract intake. DocuSign ships a single signature scope with no read-only variant — the token could send envelopes; our code never does. Cannot write: we never create, send, or void an envelope.
-
Dropbox Read only
Reads file listings and file content from the folder the customer connects for contract intake (files.metadata.read + files.content.read). Cannot write, move, or delete any Dropbox file.
-
Microsoft Dynamics 365 Business Central Reads only; the grant would allow writes
Lists your Business Central environments and companies so you can pick one, then reads only that company: its chart of accounts, general ledger entries with their dimensions, general ledger setup, company information, and aged receivables and payables, for the Financial Analyst; and its open sales invoices, credit memos, customers (name and email address), and unposted customer payment journal lines, for the AR Cash Collector. Ataski issues read requests only — note that Microsoft offers no read-only Business Central scope, so the permission you grant is technically read and write, limited to what your own Business Central user may do.
-
Freshdesk Can write
Posts the replies you approve onto the original Freshdesk ticket, and — only when you run the help-center import — reads your published Solutions articles. It uses an agent's API key from that agent's Freshdesk profile; Freshdesk keys carry the agent's full permissions, so the key can do more than Ataski uses. Ataski never edits, closes, or deletes a ticket or an article.
-
Front Can write
Posts the replies you approve onto the original Front conversation, using an API token you create in Front (Settings → Developers → API tokens) with the Shared resources scope. The token can do whatever its Front permissions allow; the only call Ataski makes with it is sending a reply to a conversation Ataski received. New messages reach Ataski through a Front webhook signed with a secret you paste.
-
Gmail (operator mailbox) Can write
Sends approved outreach and dunning email from the operator's own Gmail mailbox. The base connection requests only gmail.send — a send-only scope that cannot read the mailbox. Two optional feature grants ask for a separate re-consent: draft delivery (gmail.compose / gmail.modify, to place drafts in the operator's Drafts folder) and duplicate-contact checking (gmail.metadata / gmail.readonly, used for thread metadata lookups — message bodies are never read). Nothing sends without the operator's approval queue.
-
Google Calendar Read only
Reads free/busy availability windows only (calendar.freebusy) to propose meeting slots — never event titles, attendees, or descriptions. Cannot create or modify any calendar event.
-
Google Drive Read only
Reads the folder the customer connects for contract intake (drive.readonly). Cannot create, modify, or delete any Drive file.
-
Gusto Read only
Reads your company, its employees, and their compensation (companies:read, employees:read, compensations:read) so the Financial Analyst's headcount register can show staff and pay. Gusto assigns these scopes to the Ataski app when it reviews the app, so the consent screen lists none; every one is a read scope. Cannot create or change employees, and cannot run payroll.
-
Help Scout Can write
Posts the replies you approve back into the original Help Scout conversation. Help Scout has no OAuth scopes: the app acts with the Mailbox API access of the Help Scout user who connects it, so the permission is broader than what Ataski uses — the only call made with it is posting a reply to a conversation Ataski received. New conversations reach Ataski through a Help Scout webhook you create, signed with a secret you paste.
-
HubSpot Read only
Reads deals, contacts, and companies. Cannot create or modify any HubSpot record — CRM write-back runs on Salesforce only. A workspace that pasted a HubSpot private-app token instead of connecting is read through that token the same way: it carries the scopes ticked when the private app was created, and Ataski still only reads.
-
Intercom Can write
Reads contacts, companies, and conversations for account-health and renewal signals, reads your help-center articles when you run the import, and posts the replies you approve as an admin reply on the original conversation. It uses the access token of an app you create in the Intercom Developer Hub; what that token may do is set by the app's permissions in Intercom. Ataski never edits or deletes contacts, companies, or articles.
-
Linear (legacy) Reads only; the grant would allow writes
No longer accepted for new connections. A Linear connection made before 2026-09-07 is still read once a week for the operations digest: the list of teams, each team's active cycle, and that cycle's issues, through Linear's GraphQL API with the personal API key that was pasted in. Ataski issues read requests only — note that a Linear personal API key carries its creator's full access unless it was restricted when created, so the permission is technically read and write. Delete the key in Linear to end it.
-
Mixpanel Read only
Reads, per customer, the last event date, weekly active users, and event volumes through Mixpanel's Query API, for the Renewal Manager's account-health signals, using a service account (username and secret) you create in your project. Give it the Consumer role, which can only read; Ataski never sends events or changes the project.
-
NetSuite Read only
Reads GL transactions, AR aging, AP aging, accrual snapshots, and subsidiary roster over SuiteTalk REST. New connections sign in with OAuth 2.0 (authorization code + PKCE, scope rest_webservices) through an Integration record the customer creates in their own NetSuite account; existing connections may keep Token-Based Authentication (TBA) — an OAuth 1.0a-flavoured credential the customer generates inside their NetSuite UI (Setup → Users/Roles → Access Tokens). Either way the access is bounded by a read-only NetSuite role of the customer's choice, and Ataski only issues read queries. Cannot post journal entries, modify records, or close accounting periods.
-
Notion Can write
Uses an internal-integration token you create in Notion. A Notion integration sees only the pages and databases you share with it. Ataski reads the pages you share for the Support knowledge base, and adds one new page per delivered contract brief or financial digest to the database you choose. It never edits or deletes an existing page; what the token may do inside shared pages is set by the capabilities you tick on the integration in Notion.
-
1C:Enterprise Read only
Reads the chart of accounts, accounting-register turnovers and balances, and receivables from your 1C:Enterprise base through its published OData interface, for the Financial Analyst and the AR Cash Collector. It signs in as a 1C user you create, so access is bounded by that user's rights in 1C — give it read-only rights. Ataski only issues read queries and never posts a document or changes the base.
-
Microsoft Outlook / Microsoft 365 Can write
One Microsoft consent covers document intake (Files.Read.All), calendar availability (Calendars.Read), reply detection (Mail.Read), and sending approved outreach and dunning email from the operator's own Outlook mailbox (Mail.Send — the write surface). A connection granted without Mail.Send is never treated as send-capable, and nothing sends without the operator's approval queue.
-
Paddle Read only
Reads subscriptions and transactions so the Financial Analyst can build recurring revenue, with an API key you create in Paddle. What the key may do is set in Paddle; Ataski issues read requests only and never changes a subscription or a transaction.
-
Pipedrive Reads only; the grant would allow writes
Reads deals and the organizations they belong to, for renewal dates and the Financial Analyst's pipeline, and searches people and organizations so the Outbound Sales Assistant does not contact someone already in your CRM. Uses your company domain and a personal API token you paste in. Ataski issues read requests only — note that Pipedrive offers no read-only API token: the token carries the full access of the user it belongs to, so the permission is technically read and write.
-
PostHog Read only
Reads, per customer, recent product events and the date of last activity for the Renewal Manager's account-health signals, with a personal API key and your region (US or EU). PostHog lets you limit a personal key's scopes, and read scopes are enough; Ataski never sends events or changes your project.
-
Postmark (your own account) Reads only; the grant would allow writes
Reads bounces, sent-message records, opens, and clicks from your own Postmark server so the Renewal Manager can spot customers whose mail stopped landing, with a Server API token you paste in (Postmark → Servers → your server → API Tokens). Ataski issues read requests only — note that Postmark offers no read-only server token: the token could also send mail from that server, so it is technically read and write. This is separate from Ataski's own Postmark account, listed as a sub-processor.
-
QuickBooks Online Reads only; the grant would allow writes
Reads chart of accounts, P&L, balance sheet, cash flow, invoices, customers, and vendors. Ataski issues read requests only — note that Intuit offers no read-only accounting scope, so the permission you grant is technically read and write.
-
Recall.ai Can write
Creates bots that join scheduled meetings and reads back their transcripts and engagement metadata. Write surface is limited to dispatching the bot; cannot mutate any calendar, video-conference, or meeting-host data.
-
Recurly Read only
Reads plans, subscriptions, and invoices so the Financial Analyst can build recurring revenue, with a private API key you create in Recurly. What the key may do is set in Recurly; Ataski issues read requests only and never changes a subscription or an invoice.
-
Rippling Read only
Reads workers, users, and compensation (workers.read, users.read, compensations.read) so the Financial Analyst's headcount register can show staff and pay, with an API token you create in Rippling (Tools → Developer → API Tokens). The token sees only what its creator can see, and every scope asked for is a read scope. Ataski never changes a Rippling record.
-
Sage Intacct Read only
Reads GL accounts, journal entries, AR/AP, and trial balance through Sage Intacct's Web Services, using Web Services credentials the customer provisions: a sender ID the customer obtains from Sage and authorises in their own company, and a read-only Web Services user the customer creates in their Sage Intacct admin. No OAuth sign-in is involved. Cannot post journal entries or modify any Sage Intacct data.
-
Salesforce Can write
Reads accounts, contacts, opportunities, and activity for renewal and outbound signals, and writes back CRM activity the customer switches on. Salesforce's api scope is the full REST API — Salesforce ships no narrower data scope, so we disclose it as write-capable rather than calling the connector read-only. refresh_token / offline_access only keep the grant alive.
-
Segment Read only
Reads, per customer, recent events from the Segment Profile API (part of Segment's paid profiles plan, on your own subscription) for the Renewal Manager's account-health signals, with a Profile API access token. The Profile API only reads profiles; Ataski never sends events.
-
Shopify Read only
Reads your store's orders (read_orders; add read_all_orders, or Shopify returns only the last 60 days) so the Financial Analyst can model order value, repeat purchases, and growth. Uses the Admin API token of a custom app you create in your store; the token can do only what the scopes you tick allow, and both scopes asked for are read scopes. Ataski never creates, changes, or refunds an order.
-
Slack (Ataski app) Can write
Installs the Ataski bot in your Slack workspace. It posts notices, digests, and answers (chat:write), opens a direct message with a teammate (im:write) after looking that one person up by email (users:read.email — Slack requires users:read alongside it; the directory is never listed), registers the /ataski command (commands), and reads only messages addressed to the bot: @mentions (app_mentions:read) and direct messages to it (im:history, im:read). It cannot read channels where it is not mentioned, and cannot change workspace settings or members.
-
Slack incoming webhook Can write
A Slack Incoming Webhook URL you create in Slack and paste in (only hooks.slack.com addresses are accepted). It can post messages into the one channel it was created for — contract briefs, obligation reminders, financial report notices and accounts-receivable alerts — and nothing else: it cannot read any message, list channels, or see who is in the workspace. Stored encrypted; delete the webhook in Slack to revoke it.
-
SMTP mailbox Can write
Sends the collection emails the AR Cash Collector prepares from your own mailbox (Yahoo, Apple iCloud, Fastmail, or your own server), signing in with the username and password you enter. Ataski connects only over SMTP, which can send mail but cannot read, move, or delete it. On your own server that password may be the real mailbox password — use an app-specific password wherever your provider offers one. Stored encrypted.
-
Stripe Read only
Reads subscriptions, invoices, customers, refunds, and payment intents via a customer-provided Restricted API key. Cannot create, modify, or refund charges.
-
Microsoft Teams webhooks Can write
Two Microsoft Teams links, both created by you inside Teams and both limited to one channel. An Incoming Webhook URL (a *.webhook.office.com or Workflows *.logic.azure.com / *.environment.api.powerplatform.com address) lets Ataski post obligation reminders and accounts-receivable alerts into that channel. An Outgoing Webhook lets the Financial Analyst answer when someone @mentions it there: Teams sends only that message, signed with a secret Ataski stores encrypted and checks, and the answer goes back as the reply. Neither can read other messages, list teams or members, or act anywhere else in Microsoft 365.
-
Xero Read only
Reads the P&L, the balance sheet, the chart of accounts, tracking categories, and organisation settings, plus the source documents behind them — invoices, bills, credit notes, bank transactions, and manual journals — which is how the posting-level detail is reconstructed, and the customer and supplier contact records those documents are addressed to, which is how a receivable finds a billing email. Each of those is a SEPARATE granular scope since 2026-09-08: the two broad names Xero deprecated (all transactions, all reports) are no longer requested. Xero's general-ledger feed itself (its Journals endpoint) is sold only from Xero's Advanced app tier and is NOT read, so entries Xero posts on its own — payroll runs, fixed-asset depreciation, tracked-inventory cost of sales, currency revaluation, the year-end close — do not appear in that detail. The three payroll read scopes are OPTIONAL and off by default: they are requested only when the operator asks the headcount register to read staff and per-person pay out of Xero Payroll, and a workspace that never asks holds a token without them. Cannot create or modify any Xero data, payroll included.
-
Yandex Disk Read only
Reads PDF, Word, and Excel files in the Yandex Disk folder you connect for contract intake (cloud_api:disk.read). Cannot create, modify, or delete any file on your Disk.
-
Zendesk Can write
Reads tickets and their history for account-health and renewal signals, reads published Help Center articles and macros when you run the import, and posts the replies you approve as a public comment on the original ticket. It uses an OAuth access token from an OAuth client you create in Zendesk Admin Center; the token's scope is set there, so Ataski records it as unknown rather than guessing. The help-center import instead signs in with your account email and a Zendesk API token, which carries that user's own permissions. Ataski never deletes tickets or changes users, groups, or settings.
-
Zoho CRM + Zoho Books Read only
Reads CRM contacts, accounts, and deal stages plus Zoho Books invoices, contacts, and organisation settings for the renewal book, and the Zoho Books chart of accounts, account transactions, and open bills (supplier, balance, due date, payment terms) for the Financial Analyst. Every requested scope is a .READ scope — cannot create or modify any Zoho record.
Penetration testing
What an outside tester has found.
No external penetration test has been performed yet. The automated security checks listed above run on every change, but they are not a substitute for one.
Contract
The documents your counsel will ask for.
All of them are public and versioned. The DPA is the GDPR Article 28 contract; under the CCPA we act as a Service Provider and do not operate as a data broker. The sub-processor list carries a fingerprint recorded on every signed DPA, so you can prove which version you agreed to.
Your rights
Export, correct, delete.
Under GDPR and UK GDPR you can ask for a copy of your data, a correction, or deletion. Export is self-service from your dashboard. Deleting a workspace (Settings → Delete workspace, or email security@ataski.com) starts a 30-day grace period during which you can change your mind; the purge then runs and completes within 48 hours across our database and file storage. Copies that vendors keep on their own timers expire under each vendor's retention terms, listed in the Data Retention Policy.
Company information
Who operates Ataski.
GlobalDeal Inc.
Ataski is operated by GlobalDeal Inc., a Delaware C-Corporation. Mailing address:
GlobalDeal Inc.1720 W Ball Rd, Ste 4B #136
Anaheim, CA 92804
United States
Security, privacy and procurement: security@ataski.com
Questions
Security reviews and questions.
Email security@ataski.com for security questionnaires, the DPA, vulnerability reports or anything on this page. We acknowledge within two business days and send a written answer, usually within a week.