Munshi uses artificial intelligence to turn what you say into a structured case record. This page explains every step, so you know exactly what happens to your recordings. It supplements our Privacy Policy.
The pipeline
- Record — you tap the mic; the app records mono audio (up to 3 minutes per note). It never records unless you press the mic. The recording is saved on your phone first so it is never lost, even offline.
- Upload — when you are online, the audio is uploaded over an encrypted connection to our servers.
- Transcribe — a third-party AI speech service converts the audio to text, with hints for legal vocabulary (peshi, tareekh, zamanat, PPC/CrPC sections, court names).
- Structure — a third-party AI language service reads the transcript and fills a strict set of fields (client, parties, court, judge, FIR, sections, date, time, purpose, summary). If it is not sure about a field, it leaves it empty or flags it.
- You review — the draft appears in the app. Unsure fields are amber and marked "check". You can edit by tapping or by voice.
- You approve — only then is the case saved to your diary and reminders scheduled. AI never saves anything silently.
Third-party AI services
Munshi uses third-party AI services for these steps, so your voice recordings, transcripts and relevant diary text are sent to those services for processing.
| Step | Type of service | What it receives |
|---|---|---|
| Speech-to-text | Third-party AI speech service | Your audio |
| Structuring, summaries, Ask Munshi | Third-party AI language service | Your transcript and relevant diary text |
We choose providers for accuracy and privacy, and may change them. The types of provider we use are listed on the sub-processors page.
No training on your data
We use these services through business APIs, whose terms state that data submitted through the API is not used to train their models by default. We do not train our own models on your recordings. We do keep the text of corrections you make to drafts (without audio) to improve our prompts and accuracy.
Who can hear or read it
No person at Munshi listens to your recordings or reads your transcripts as part of normal operation. If you ask support to investigate a specific problem, we ask your permission first and access only what is needed.
Retention
Audio is deleted automatically 30 days after recording by default; you can change this period in More → Privacy & security. Transcripts and drafts stay with the case until you delete it or your account. See the data retention schedule.
Ask Munshi
When you ask a question, Munshi searches only your own diary (and your chamber’s shared data, if any) and uses the language model to phrase the answer. It does not search the internet, other users’ data or legal databases, and it does not give legal advice.
Accuracy and responsibility
Speech recognition and AI can make mistakes, especially in noisy corridors. Always check amber fields and confirm dates against official court records. You are responsible for entries you approve.
Contact
Questions about AI processing: support@digitalmunshi.com.pk