Trust & data
Your calls stay yours
Dialogist runs its own and open-weight models on its own servers in the EU, or on yours. No caller audio or text goes to a third-party AI provider. Here is where your data lives, who can see it and what an AI voice agent is allowed to do.
Where your data lives
EU cloud
The Dialogist cloud is hosted in the EU.
Your region
Where data has to stay in another market, we set up an isolated regional cluster on request, with its own voice and knowledge engines and its own storage.
Your servers
On-premise, the whole platform runs inside your network: control layer, voice engine, knowledge engine and every model. Air-gapped installations are supported.
Who processes it
Models we run ourselves
Speech recognition (Oasis Speech), the language model, speech synthesis, turn-taking and knowledge search all run on servers we operate, or on yours.
Our own and open-weight models
Nothing in a call depends on a third-party AI service.
No resale of AI
We do not resell a third-party AI service, and the consulting team is our own.
Who can see it
Roles
Owner, Admin and Member. Members have read-only access to the voice platform.
Organizations kept apart
Agencies and resellers can run several client organizations from one login; each organization’s agents, knowledge, numbers, data, plan and limits stay separate.
Fully isolated instances
For organizations whose rules do not allow shared infrastructure: a dedicated installation, separated from every other customer.
Signing in
Password or email link
With optional two-factor authentication.
Enterprise SSO with SAML
Planned.
Recordings and transcripts
Recording is your switch
Calls land in the journal with a transcript and a summary; the recording is kept when you switch recording on.
The caller is told
The greeting can announce that the caller is speaking to an AI assistant and that the call is recorded. You choose the wording.
Structured results
Each call produces the fields you define, so your systems receive what was agreed rather than the whole conversation.
What the agent is allowed to do
Tools are declared per agent
An agent can call only the endpoints you have given it, with the call and caller values filled in for it.
Connectors have scopes
When you connect Salesforce, Creatio or Zendesk, you choose what the agent may read and what it may write.
Thinking ahead cannot act
When the agent prepares a reply before the caller’s pause has ended, that preparation can only read: it writes nothing to the conversation, calls no tool and cannot end or transfer the call. If the caller says something else, the agent throws it away.
People decide what matters
In candidate calls the agent records answers and books interviews; it does not reject anyone. Disputes, hardship and anything outside your rules go to a person, with the conversation so far.
Lines read in full
For scripts with required wording, you can set an agent to read its lines to the end without yielding to interruptions.
Tested before it talks
Rehearsed in Polygon
You can rehearse every change to an agent in Polygon, our conversation test bench, before a caller hears it.
Hard gates
A forbidden phrase in the agent’s speech, a safety violation or an invalid tool call that reached the caller fails a conversation whatever its score.
Report a vulnerability
Report a vulnerability
Write to security@dialogist.cc. Our disclosure details are in security.txt.
Status
Live availability of the platform.