A voice agent your agency runs in-house
Dograh is the self-hostable voice agent platform with an open-source workflow builder. Host the platform in your data center or government cloud, and add open-weight models when citizen audio must stay inside the boundary.
The calls that clog a public line
Pick up every line at once
Inbound helplinethe number citizens already call
Routine questions about a form or an appointment get an answer in 70+ languages, at 2 a.m. as well as noon.
→ Let the agent wait for slow or pausing speakers
Answer from the citizen's file
Pre-call fetchthe record loads before the agent speaks
The agent verifies the caller first, then reads the case or permit status from your own systems.
→ Share nothing from the file before the identity check
Send judgment calls to staff
Call transfera person picks up the call
When a case needs a decision or a human voice, the agent transfers the caller to your staff.
→ Transfer to a person works on Twilio
Decide how far inside the boundary it runs
Your own retention and access rules decide who opens them.
Only models you host keep the audio inside. A hosted model provider still hears the call.
Answers reach your case files after each call.
Check these before citizens call in
- Caller verified firstThe agent confirms an approved detail before it reads anything from a citizen's file.
- Languages testedEach language you promise was tried by a native speaker over a real phone line.
- Unsure language handledIf the agent can't tell which language it heard, it asks, and transfers when it still can't help.
- Transfer tested by phoneBrowser test calls cannot transfer, so try the handoff to staff on a real call.
- Model endpoints checkedEvery model URL points inside your boundary if citizen audio must not leave.
Built for an agency security review
Runs where you accredit
Install Dograh in your own data center or government cloud, behind your firewall.
Code your team can audit
The platform is open source, so security staff read how a call is handled instead of taking a vendor's word.
Audio stays inside, if you choose
Pair the platform with open-weight models you host, and no call audio reaches an outside provider.
No meter on busy lines
Self-hosted, there is no per-minute platform fee. Servers and phone lines still cost money.
What changes when the agency hosts it
| What you get | Dograh | Hosted voice platforms |
|---|---|---|
| Open-source code your security team can audit (BSD-2) | ✓ | ✕ |
| Runs in your own data center or government cloud | ✓ | ✕ |
| Recordings and transcripts kept in your storage | ✓ | ✕ |
| No per-minute platform fee when self-hosted | ✓ | ✕ |
Six citizen calls, step by step
Where does citizen call data live?+×
Recordings and transcripts stay on your servers when you self-host. The live audio stays inside only if you host the models too.
Can we run our own speech and language models?+×
Yes. Use open-weight models you host, keys to models your agency already approves, or Dograh's own stack.
How many languages can the agent handle?+×
70+ languages, so callers with limited English reach service in their own language.
Can the agent transfer a citizen to staff?+×
Yes, on Twilio. The agent hands the call to a person when it needs judgment. Test the transfer on a real phone call.
Is there a per-minute fee?+×
Not on the self-hosted platform. You pay for your servers and phone lines instead.
Is the platform open source?+×
Yes. Dograh is BSD-2 licensed, so your security team can read exactly how the agent handles a call.
Start with your busiest helpline.
Self-host in one command, or start on Dograh Cloud.
$ curl … start_docker.sh && ./start_docker.sh