Instalment calls that keep spending data in-house
Dograh is the self-hostable voice agent platform with an open-source workflow builder. Remind, recover and set up catch-up plans in your own name, with every figure read live from your loan system and borrower data kept on servers you run.
Before the pull, after a miss, back on track
A call a day or two before the pull
Campaigna CSV list the agent works through
Upload the instalments due soon, and each column becomes a value the agent can say once the borrower passes the check.
→ Split the list by time zone
The live balance, not the morning file
Pre-Call Data Fetcha lookup as the call connects
Your loan system answers with the current amount and paid status, so a borrower who already paid is thanked, not chased.
→ Keep the uploaded list fresh as a fallback
Only the plans your system offers
HTTP API toola request to your system mid-call
The agent reads the catch-up options this borrower qualifies for, repeats the chosen dates back, and a second tool saves the plan.
→ Hardship and disputes go to a person
Your figures, your name, your storage
Switch off Allow Greeting Interruption so the lender's name always plays in full
The agent says no date, fee or plan that your system did not return
With your own model keys the provider still hears the audio, so host the models too
Test these before a borrower hears the agent
- Amount waits for the checkA wrong-person test call hears only the lender's name and a callback number.
- Payment by secure linkYour system texts or emails its own payment link, and no card number is spoken on the call.
- Plan options match the APIEvery option the agent reads is one your system returned for that borrower.
- Call limits kept in your listWeekly limits per borrower, such as state caps, are applied before you upload each list.
- Retries chosen on purposeSet 0 to 10 retries and pick which of busy, no answer or voicemail gets another try.
Why BNPL teams run it themselves
Purchase history stays put
Self-host with open-source models, and what borrowers bought never reaches an outside AI vendor.
Hardship reaches a person
The agent logs the reason, then rings your hardship team and brings the borrower in once someone answers.
Each call leaves proof
Recordings, transcripts and per-step QA scores show the amount came only after the identity check.
Outcomes land in your loan system
A Webhook node sends the disposition, any promise-to-pay date and the recording link after every call.
Every BNPL call, explained in full
Do pay-in-four calls need credit-card disclosures?+×
Generally not. An interest-free plan with four or fewer payments generally sits outside Regulation Z's creditor definition. Confirm your own plans with counsel.
Does the FDCPA cover a lender collecting its own instalments?+×
Generally not, while you collect under your own name. Calling as what sounds like a third-party collector can change that, so fix the greeting wording.
Do AI reminder calls to mobiles need consent?+×
Generally yes. The FCC treats AI voices as artificial, so calls need prior express consent, often the number the borrower gave for the loan. Any offer needs written consent.
Can borrowers pay during the call?+×
Not by voice. The agent asks your system to text or email a secure link, and the borrower pays on your own payment page.
Where does borrower data live?+×
Self-hosted, recordings, transcripts and traces sit in your storage. Live audio stays inside too when you host open-weight speech and language models next to it.
What does it cost to run?+×
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.
Start with the reminder before the next pull.
Self-host in one command, or start on Dograh Cloud.
$ curl … start_docker.sh && ./start_docker.sh