Chat Cactus
Sign in
Security

Security, stated plainly

No badges we haven't earned, no claims the code doesn't back. Here is exactly how your data is handled — and what we deliberately don't do with it.

Start freeTalk to us

Overview

Chat Cactus stores your knowledge documents, conversations and leads so your team can work with them; sends conversation context to AI providers for inference only; and keeps the most sensitive material — third-party credentials and payout details — encrypted at the application layer. Payments never touch our servers. What follows is the specific version of that sentence.

Encryption

  • In transit: all traffic is served over HTTPS, with HTTP Strict Transport Security (6-month max-age, subdomains included).
  • Integration credentials at rest: OAuth refresh tokens, Twilio auth tokens and bot tokens are AES-256-GCM encrypted before they reach the database; referral payout details get the same treatment.
  • Passwords: bcrypt-hashed, never stored or logged in plaintext.
  • API keys: stored only as SHA-256 hashes — the full key is shown once at creation and cannot be retrieved afterwards, only revoked.

How your data flows

  1. 1
    A customer writes on any channel
    Widget, email, SMS, WhatsApp, Telegram, Slack, Discord, Messenger, phone, or your own webhook. Push channels are signature-verified; the customer’s message is stored as part of the conversation.
  2. 2
    Relevant passages are retrieved
    The question is matched against your uploaded documents’ embeddings; only the few most relevant passages join the prompt — not your whole knowledge base.
  3. 3
    One inference call answers it
    The prompt (persona, retrieved passages, recent conversation turns) goes to the model provider you chose — OpenAI or Anthropic — for inference only. The reply and its citations come back.
  4. 4
    The reply is delivered and recorded
    Sent back on the customer’s channel and stored in the conversation transcript your team sees in the dashboard, with per-message model cost logged against your workspace’s spend ceiling.

Retention and deletion

  • Conversation transcripts are retained so your team can work from them; deleting an agent deletes its conversations.
  • Knowledge documents: deleting one immediately deletes its extracted text, chunks and embeddings (database-level cascade).
  • Leads can be deleted individually from the dashboard at any time.
  • Audio is never retained: no call recordings, no stored microphone clips, no saved speech output — transcripts are text.
  • Full workspace or account deletion is handled by support on request today; self-serve deletion is on the roadmap.

Subprocessors

Third-party services your data can reach, and why:

ServicePurpose
OpenAIAI inference: chat models, document embeddings, widget speech-to-text and text-to-speech
AnthropicAI inference: Claude chat models (Pro plans)
StripeSubscriptions and payments (hosted checkout — card data never touches Chat Cactus)
TwilioSMS, WhatsApp and inbound voice delivery (phone speech uses Twilio’s recognition and Amazon Polly voices)
GoogleGmail OAuth and sending/receiving mail for connected inboxes; Google Analytics on the marketing site
MetaFacebook Messenger message delivery for connected Pages
Telegram / Slack / DiscordMessage delivery for connected bots
Email provider (Resend or SMTP)Transactional email: invitations, password resets, escalation alerts
SentryError monitoring
Amazon Web ServicesApplication hosting

Channel providers only receive the messages that must transit them to reach your customer. AI providers receive conversation context for inference only.

Access control

  • Workspace roles — Owner, Admin, Member — gate billing, member management, integrations, API keys and document deletion.
  • The public widget respects per-agent domain allow-lists and per-visitor rate limits; the REST API is rate-limited per key.
  • Monthly message quotas and a per-workspace AI spend ceiling cap blast radius — as spend nears the ceiling, requests automatically downgrade to a cheaper model.

Compliance status — accurate as of September 2026

We are not SOC 2 certified and do not imply otherwise. If your procurement or legal process requires a data-processing agreement or specific documentation, contact us — we'll state plainly what we can provide today rather than badge-collect.

Reporting a security issue

Email [email protected] with "SECURITY" in the subject. Reports go to the engineers who own the code, and we'll acknowledge yours quickly.

Frequently Asked Questions

Do you train AI models on my data?

No. Your documents and conversations are sent to model providers for inference only — answering the question in front of them. There is no fine-tuning or training pipeline of any kind in the platform.

Do you store my customers’ card details?

We never see them. Checkout and plan changes run entirely on Stripe-hosted pages; our database stores only Stripe’s opaque identifiers.

Are voice calls or microphone audio stored?

No. Phone calls are never recorded, widget audio is transcribed and discarded, and generated speech is streamed, not saved. Text transcripts are what your team sees.

Are you SOC 2 certified?

Not today, and we won’t imply otherwise. If your procurement process needs specific documentation, contact us and we’ll tell you plainly what we can and cannot provide right now.

How do I get my data deleted?

Knowledge documents and their embeddings delete immediately from the dashboard; leads and agents likewise. For full workspace or account deletion, contact support and we’ll process it — self-serve deletion is on the roadmap.

Keep exploring

Privacy policy

The legal counterpart to this page.

Learn more
Platform features

What all of this security protects.

Learn more
Documentation

API keys, webhooks and signatures in detail.

Learn more

Built carefully. Described honestly.

10 minutes from sign-up to your first answered customer message. Free forever plan, no card required to build.

Start freeSee pricing