Every caller hears the AI notice first
Dograh is the self-hostable voice agent platform with an open-source workflow builder. Put the AI notice in a greeting callers cannot cut short, check sampled calls with QA and keep every trace on EU servers you run.
Notice, recording line, then a check
The AI notice comes first
Greeting Textthe opening line, played word for word
Write the notice into the greeting, or play a clip in a real voice, so callers hear it before any question.
→ Switch greeting interruption off
Recording gets its own sentence
Recording noticea rule apart from the AI notice
Place it after the AI notice, so callers can tell the two apart and you can edit each alone.
→ Plan for callers who refuse consent
Prove the notice was heard
QAscoring that runs after a call ends
Run QA on the calls you sample, and keep recordings and traces as proof of what each caller heard.
→ Hear a batch of real calls before widening
Fixed words up front, evidence kept in the EU
Bilingual lines need the notice in both languages, since it plays before the caller speaks.
A fixed Transition Speech line beats a prompt, which the model rewords on each call.
Audio stays in the EU only when the models run there too.
Check these before the first live call
- Notice survives an interruptionTalk over the greeting on a test call, then confirm the whole notice played.
- Nothing asked before the noticeAsking for details ahead of the notice generally misses the first-interaction timing.
- Each language coveredInbound lines open in every likely language, and campaigns set the notice per contact.
- Model endpoints checkedEvery model runs on your servers or with a provider whose EU processing you checked.
- Role confirmed with counselA business running its own branded agent is generally the provider, so confirm yours with counsel.
Control the words, the servers and the record
EU servers you pick
Self-host Dograh with Docker in an EU data centre, on machines your team manages.
Approved wording, every call
The model never rewrites the greeting, so the notice you sign off is the notice each caller hears.
Swap models, keep the agent
Bring your own keys or move to local models without rebuilding a single step of the flow.
Evidence on your side
Recordings and full traces of every greeting sit in your own storage for any regulator who asks.
What running it yourself adds in the EU
| What you get | Dograh | Closed hosted voice platforms |
|---|---|---|
| Whole platform self-hosted in an EU data centre you choose | ✓ | ✕ |
| Open-source code your legal and security teams can audit | ✓ | ✕ |
| No per-minute platform fee on your own install | ✓ | ✕ |
The AI Act for phone agents, part by part
Do voice agents in the EU have to say they are an AI?+×
Generally yes. Article 50 asks for the notice at the latest at the first interaction, unless it is obvious, and that exception is read narrowly.
How do I make sure callers hear the whole notice?+×
Put it in the Greeting Text on the Start Call node and switch Allow Greeting Interruption off. Caller speech is then held until the greeting ends.
Does self-hosting make the agent compliant?+×
Not by itself. EU servers decide where call data sits. The AI notice is about what the agent says, so it still belongs in the greeting.
Did the AI Omnibus delay the AI notice?+×
No. It moved the high-risk deadlines but left the duty to tell callers they are talking to an AI as it was.
Does the AI notice cover call recording?+×
No. Recording falls under the ePrivacy Directive, national law and the GDPR, so it generally needs its own sentence after the AI notice.
What does running it cost?+×
Self-hosted Dograh has no per-minute platform fee, but servers, model keys and phone minutes still cost money. Dograh Cloud charges a 1¢-a-minute platform fee.
Write the AI notice into your greeting first.
Self-host in one command, or start on Dograh Cloud.
$ curl … start_docker.sh && ./start_docker.sh