Enterprise security
Security controls for an AI chatbot your whole company works on
Decide who can change which agent, sign your team in with your identity provider, see who changed what, and limit where each agent answers and who can talk to it. This page lists what intoCHAT does today, and what it doesn't.
An AI agent touches your content, your customers and your systems
A support agent reads your documents, talks to your customers, can call your APIs and is edited by several people. Before a security review signs it off, someone has to answer: who can change the instructions, who can connect a helpdesk or create an API key, what happens when someone leaves, and where do visitors' messages go? A chatbot that only has one shared login and no record of changes makes those questions hard to answer. intoCHAT gives you roles, single sign-on, an audit log and per-agent controls, and its documentation states the limits plainly so your review can rely on it.
Five layers of control
Each one is documented in detail, including what it does not cover.
Roles and per-agent access
Give each person Admin, Editor or Viewer, and limit editors and viewers to selected agents. On Enterprise, build custom roles with None, View, Edit or Manage for each area of the account.
Single sign-on with SAML 2.0
On Enterprise, your team signs in through Okta, Microsoft Entra ID, Google Workspace or any SAML 2.0 identity provider, joins the team automatically with the role you choose, and can be required to use SSO.
An audit log of changes
See who created, changed or deleted agents, sources, actions, webhooks, integrations, API keys and team members, and who downloaded exports. 90 days on Pro, 365 days on Enterprise.
Controls on every agent
Limit an agent to your own domains, make it private, block countries, cap messages per visitor and per month, pause spam, and let only visitors your site has signed in chat.
Honest about the limits
intoCHAT holds no SOC 2, ISO 27001 or HIPAA certification, and says so. The docs list what each feature leaves out, such as SCIM or IdP-initiated sign-in, so nothing is a surprise later.
How to lock down an account
Give people the least access they need
Under Settings, Team, invite people as Viewer or Editor rather than Admin, and choose Selected agents for anyone who works on only some agents. On Enterprise, create a custom role for jobs the built-in roles don't fit.
Connect your identity provider
On Enterprise, open Settings, Single sign-on, add your IdP's metadata, verify your email domains with a DNS TXT record and click Test sign-in. Then switch on Require SSO for members with these domains.
Decide where each agent answers
On the Share tab, list your allowed domains, switch on Private agent while an agent isn't ready, and set messages per visitor, a monthly cap and blocked countries under Limits and access.
Know who is chatting
Add identity verification so your server signs each signed-in user's id. Switch on Only verified visitors can chat for agents that should serve logged-in customers only.
Security details
Custom roles on Enterprise
Up to 20 roles, each with a level per area: agent settings, knowledge, actions and procedures, leads and contacts, conversations and live chat, analytics, integrations and webhooks, team, billing and API keys. Changes apply on the next click.
Verified domains and Require SSO
SSO works only for addresses on domains you verified, and a domain can belong to one account only. With SSO required, password and Google sign-in are refused for those addresses. The account owner is never locked out.
What the audit log never records
No passwords, API keys, tokens or other secrets, no message content and no visitor details, and nothing people only looked at. IP addresses are cut to the /24 or /48 network. Filter by action, member, agent and date, and export up to 10,000 rows as CSV.
Two-factor sign-in
Each person can protect their email-and-password sign-in with a 6-digit code from an authenticator app, with 10 single-use recovery codes. intoCHAT stores only hashes of the recovery codes.
Scoped API keys and connectors
From Starter, API keys can be limited to the scopes they need, up to 20 active keys per account. MCP connectors such as Claude.ai or ChatGPT can be allowed read-only access, and the person's role and agents still apply.
Approval before sensitive actions
On Pro and Enterprise, an API action or procedure step can require team approval: the call waits until someone approves or rejects it, and approving runs it once with exactly the inputs that were waiting.
Built-in rules against prompt injection
The agent treats your documents, quoted material and tool results as untrusted reference data, not instructions, which helps it resist attempts to change its role, reveal secrets or call tools it shouldn't.
Encrypted credentials, signed webhooks
Keys and tokens for connected services are encrypted with AES-256-GCM and never shown again, only their last four characters. Webhooks carry an X-IntoChat-Signature header so your server can check they came from intoCHAT.
Frequently asked questions
Does intoCHAT support SAML single sign-on?
Yes, on the Enterprise plan. It works with Okta, Microsoft Entra ID, Google Workspace and any identity provider that supports SAML 2.0, and requires signed assertions. Sign-in starts at intoCHAT; IdP-initiated sign-in, OpenID Connect, single logout and encrypted assertions are not supported, and there is one identity provider per account.
Is there SCIM provisioning?
No. People join your team when they first sign in with SSO, with the role you chose for new members, and roles don't come from IdP groups. Removing someone at your IdP doesn't remove them from your intoCHAT team: remove them in Settings, Team as well.
Is intoCHAT SOC 2 or ISO 27001 certified?
No. intoCHAT holds no security or data protection certification, such as SOC 2, ISO 27001 or HIPAA. If your policy requires one of these, intoCHAT doesn't meet it today. What it does offer is described on this page and in the documentation.
Is intoCHAT GDPR compliant?
We don't claim a certification. For your visitors' data intoCHAT acts as a processor on your behalf, and you can request a data processing agreement at privacy@intochat.ai. According to the privacy policy, the database is hosted in Switzerland, and some providers process data in the United States and other countries, under standard contractual clauses where the law requires them. You can delete conversations and leads at any time.
Where do visitor messages go, and are they used to train AI?
To write a reply, the visitor's message, recent history and the relevant parts of your knowledge go to the provider of the model you chose: OpenAI by default, or Anthropic or Google. OpenAI is also used to index knowledge and for background features. intoCHAT doesn't train AI models on your data or your visitors' data.
How long are conversations kept?
Until you delete them, or the agent or account. There is no automatic deletion schedule. You can delete a conversation with its files, let visitors delete their own chat, and offer temporary chats that are never saved. Ending idle conversations closes them but doesn't delete them.
Can I require two-factor authentication for my team?
Not yet. Each person turns two-factor authentication on for their own login, and the Members list shows a 2FA badge next to those who did. With SSO, multi-factor authentication is your identity provider's job, so you can require it there.
Which plan do I need?
Roles and selected-agent access come with team seats from Starter (2 seats; Pro has 5). API keys and connectors start at Starter. The audit log page and team approval start at Pro. Single sign-on, custom roles, unlimited seats, 365 days of audit log, an SLA and dedicated support are on Enterprise.
Put your AI agent through your security review
Read the docs on roles, single sign-on and the audit log, or create a free account to try the controls. For Enterprise, contact us.
Create your agent freeFree plan, no credit card needed.