Data Security with Informer GO

Informer GO is the companion app for Informer on desktop, web, and mobile: AI chat, data exploration, custom Apps, and integrations with outside tools. It adds those surfaces without adding a second security model. GO does not bypass, relax, or reimplement Informer's permissions; every request it makes goes through the Informer server, where the same scopes and access controls apply as anywhere else.

This article follows a request from the User who types it through to the AI provider and back, and shows what protects it at each stage.

Informer's security model still applies

Informer's three-tiered Data Governance framework (Functional, Content, and Data) is fully in force in GO.

• Functional level separates Normal Users, who work within their Team roles, from Super Users, who have full system access. In GO those same roles decide what a User can do: only Super Users manage AI providers and models or change system-wide AI settings, and Normal Users use AI features only within their Team roles.
• Content level governs access to content by Team role. When an AI chat reaches for Datasets, Libraries, Assistants, or Toolkits, it sees only what has been shared with that User or their Team. A Dataset the User cannot open in Informer is a Dataset the AI cannot open either.
• Data level covers Mapping Sets, Restricted Fields, Dataset filters, and row-level security, set by Team Admins, Publishers, and Data Wizards. When the AI searches or queries a Dataset on a User's behalf, it runs under that User's exact data permissions and sees only the Fields and rows they are authorized to see.

How data reaches an AI provider

Informer supports several AI providers, including OpenAI, Anthropic, Google, and self-hosted models. The path is the same for all of them: GO never talks to an AI provider directly. A message goes to the Informer server, which authenticates the request, applies permissions, and makes the provider call itself.

That one architectural choice is what the rest of this rests on:

• Provider API keys live with Informer. They are never sent to or stored on a User's device.
• Keys are encrypted at rest and decrypted only at the moment a request is made.
• The server decides exactly what goes into each AI request.
• Every request is authenticated and authorized before any data leaves for a provider.

Self-hosted models. Organizations that need complete data sovereignty can connect a model running on their own infrastructure. Requests then travel only between the Informer server and that model server, and nothing goes to a third-party AI service.

Training. The AI providers Informer supports natively do not use API data to train models. If your organization runs a self-hosted model or supplies its own API key, verify the data training settings on that provider account.

Enterprise keys. Enterprise customers use their own provider API keys, so enterprise AI usage is billed to and governed by the customer's own provider accounts.

Authentication

GO verifies who a User is before any data access happens, using the method that fits the platform:

• Desktop and web: Users sign in with their Informer credentials through GO, which supports every configured authentication method. A secure session token is issued on success.
• Mobile: GO uses an OAuth 2.0 flow designed for mobile applications. Temporary authentication data is cleaned up as soon as login completes.
• Everywhere: An expired or invalid token prompts an immediate re-authentication. No request proceeds without a valid session.

What happens during a conversation

1. The User sends a message

   Only authenticated Users with AI access can start a conversation.

2. GO builds the request

   The message goes out with the conversation's context: which Datasets, Assistants, Toolkits, and Skills are attached.

3. The server authenticates and authorizes

   Four checks run before anything else happens. The bearer token is validated, and invalid or expired tokens are rejected. Usage is checked against the plan's weekly and session budget limits, and a request that would exceed them is blocked. The plan is checked for access to the requested model's tier (Everyday, Advanced, or Strategic). Finally, the server loads only the Datasets, Libraries, Assistants, and Toolkits the User has been granted; anything that fails the access check is left out, and the AI never learns it exists.

4. The server calls the provider

   Only once every check passes does the server assemble the payload: the conversation messages, the system instructions for the Assistant or chat, the tool definitions describing what the AI may do, and only data the User is authorized to see. Entire Datasets are never sent. When the AI needs to answer a question about data, it calls a search function that queries the data on the server, under the User's permissions, and gets back only authorized results.

5. The response comes back

   The provider's answer returns through the Informer server to GO. If the AI asks to use a tool, the server runs that function under the User's permissions and hands the result back for the AI to work with. The cycle repeats until the AI produces a final answer.

Auditing

Every AI interaction is recorded in an audit log with the User, model, provider, token usage, cost, and request source, so administrators can see AI usage across the organization.

External tools

Toolkits connect the AI to outside services, and they are governed by the same permissions as any other Informer content: only Users with the right roles create or manage a Toolkit, and only Users granted access can use one. Where a Toolkit needs to authenticate to an external service, Informer supports both shared credentials and per-user authentication, where each User connects independently.

In short

Concern	How Informer GO addresses it
Who can use AI features?	Only Users granted AI access by an administrator.
What can the AI see?	Only data the User is authorized to see, enforced server-side.
Where are provider API keys stored?	Encrypted by Informer, never on a User's device.
Who can change AI configuration?	Only Super Users manage providers, models, and billing.
Are interactions logged?	Yes. Every one, with User, model, token usage, and cost.
Can external tools reach unauthorized data?	No. Tools run under the User's own permissions, and Toolkit access uses the same permissions system as other content.
Does my Dataset data go to the AI provider?	Raw data does not. The AI forms a query, the Informer server runs it, and only the results come back.
Could a provider train on my data?	The providers Informer supports natively do not train on API data. Verify the setting yourself for a self-hosted model or your own API key.

See also Data Security with Shared Assistants for keeping attached data secure when an Assistant is shared outside your deployment.