Security

Stimulus builds one record of the businesses your organisation works with from the accounts and files you choose to connect. Every layer is built so a buyer or security reviewer can check what entered the record, where each fact came from, and who approved any action.

How data enters

Stimulus reads only what your team connects or uploads. Connecting a Gmail or Microsoft Outlook mailbox, selecting the folders or labels in scope, or uploading files and exports is the consent and instruction to read that source.

Where your own record is silent, Stimulus researches public business sources and labels those findings as research so they are never confused with facts from your own record.

The record

PostgreSQL row-level security policies (FORCE ROW LEVEL SECURITY) are enforced by the database on every table in the record, scoping queries on the record to the caller's organisation and the records it is permitted to read.

The ledger is append-only. Facts are not overwritten in place: history entries are hash-chained, and facts keep both when they were true and when Stimulus learned them, so history cannot be rewritten quietly.

Every fact linked to its source

Nothing becomes a fact in your record without a link to the message, file or ledger entry it came from. Every answer and business profile opens the underlying evidence directly, and public web findings are labelled as research.

Controls for connected actions

Connected actions generally require approval. A member can choose Bypass for their own actions, subject to organisation controls that may still require approval.

AI models and your data

Stimulus uses AI models on Google Cloud Vertex AI to read, research and draft on your behalf: Google's Gemini models, and Anthropic's Claude models served through Vertex AI. Both providers are listed under Subprocessors.

  • Your data does not train AI models. Stimulus does not train models on it, and neither may its providers: Google will not use customer data to train or fine-tune any model without the customer's permission (Google Cloud Service Terms), and Anthropic may not train models on customer content (Anthropic Commercial Terms).
  • Web search keeps no logs of your queries. When Stimulus searches the web, it uses Google's Web Grounding for Enterprise, which does not log customer data (Google documentation).
  • What a provider may keep. Google may log prompts to detect abuse, and caches model responses for no longer than 24 hours (Vertex AI data retention). Stimulus has not signed a zero-data-retention agreement with any model provider.
  • Where processing happens. Model calls use Vertex AI's global endpoint, which does not fix the region where a request is processed.

Subprocessors

Third-party service providers that process data for Stimulus are listed below; some are used only when the relevant feature is enabled. Where a processing region is not pinned in code or verified in the register, it is shown as not stated.

ProviderRoleData receivedRegion
VercelApplication hosting, serverless functions, scheduled jobs, private file storage (Vercel Blob), agent workspace sandbox, and durable workflows and queuesRequest traffic (identity, session cookies, customer content, prompts and runtime logs), uploaded documents and workspace files, sandbox working state, and durable job inputs and step stateNot stated (Workflow queue host: iad1, Washington, D.C., US)
NeonManaged PostgreSQL database for the record, account and operational tables, and authentication tablesStored customer data: accounts, identity, organisation data, relationship facts and sealed OAuth tokensaws-us-east-1 (US East)
Google Cloud and GoogleVertex AI model path: Gemini models for the agent and for company enrichment, with Web Grounding for Enterprise for web search, Enterprise Knowledge Graph and Places API for company identity; Vertex gemini-embedding-001 embeddings for semantic search; company logo favicons; and OAuth sign-inConversations, prompts, tool results and workspace content read by the model; company names, domains and public web text for enrichment; company-profile text for embeddings; company domains for favicons; and sign-in identity (email, name and verified-email claim)Not stated (Vertex AI global endpoint)
AnthropicClaude language models served through Google Cloud Vertex AI, for model calls that do not name a model (the agent itself runs on Gemini)Prompts and content in the model calls Claude serves, reaching Claude through Vertex AINot stated
ComposioConnector platform for OAuth connections and tool calls to customer-connected applications (such as Gmail, Calendar, Slack, Notion, Google Drive and GitHub)OAuth tokens for connected accounts, user and organisation identifiers, tool arguments, and data returned from connected applications (email, files and messages)Not stated
ResendTransactional authentication, invitation, billing and product email deliveryRecipient email address and message content, including one-time links when applicableNot stated
SlackWebsite lead notifications when configuredLead contact information and submitted detailsNot stated
StripePayment processing and billingBilling and payment details needed to process purchases and manage subscriptionsNot stated
Microsoft (Entra ID)OAuth sign-in identity providerSign-in identity (email, name and verified-email claim)Not stated
SentryError capture and performance tracing across the web, application and agent surfacesError events, stack traces, sampled request traces (with cookies, headers, bodies, query strings, user info and AI inputs/outputs turned off and scrubbed), and agent turn telemetry (model name, tool name, token counts, cost and a hashed conversation identifier)US (us.sentry.io)
PostHogProduct analytics across the web, application and documentation surfaces, after the visitor's analytics choice (asked first in the EU, EEA, UK and Switzerland; declinable everywhere). Session replay, autocapture, heatmaps, rage clicks and dead clicks run only for signed-in members of the application, with every form input masked and agent-written text blockedPage views, page leaves, web vitals, user and organisation identifiers, a member's email and name and their organisation's name, named product events with fixed non-content properties, session recordings and clicks within the application for signed-in members (which include on-screen workspace content such as business and contact names), client IP address and user agentUS Cloud (us.i.posthog.com), reached from browsers through s.getstimulus.ai

Legal and compliance documents

Security review and vulnerability reporting

To request security documentation, discuss your organisation's requirements, or schedule a security review call, write to support@getstimulus.ai.

To report a security vulnerability, email support@getstimulus.ai. Machine-readable disclosure metadata is published under RFC 9116 at /.well-known/security.txt.