Call Transcripts

Where to find transcripts

Transcripts are saved inside your voice gateway's install directory, in a folder called transcripts:

~/ninja-talk/transcripts/

# Example listing:
2026-05-25_0102_uJrf0VGAUh1_.txt
2026-05-25_1430_Qx-DShEYW8WH.txt
2026-05-25_1815_6ht5-xbKkIKB.txt

The folder is created automatically on the first call. Each file is named with the date, time (UTC), and a short call ID so they sort chronologically.

What a transcript looks like

Each file is a clean, human-readable record of the entire conversation:

Call ID:    v3:Qx-DShEYW8WH_ZLYoWA53w...
Date:       2026-05-25 14:30:02 UTC
Direction:  outbound
From:       +16473705600
To:         +16478359044
Duration:   85s
Hangup:     normal_clearing

──────────────────────────────────────────────────

[Agent] Hi, this is Sol calling on behalf of Vello.
Is now a good time to talk?

[Caller] Yeah, what's this about?

[Agent] You have an outstanding balance of $154.56
for 1,000 SMS credits. Our records show the payment
hasn't been received yet. I'm calling to help get
this settled.

[Caller] Oh right, I forgot about that. Can you
send me a link?

[Agent] Of course. I'll send a payment link to
www.paymenttest.com right now. Is there anything
else I can help with?

[Caller] No, that's it. Thanks.

[Agent] Great, thanks for your time. Goodbye.

──────────────────────────────────────────────────
End of call

The header gives you the call metadata at a glance. The body shows every turn of the conversation with clear [Agent] and [Caller] labels. Works with any language your agent speaks.

Reading transcripts

View the latest call

ls -t ~/ninja-talk/transcripts/ | head -1 | xargs -I{} cat ~/ninja-talk/transcripts/{}

Search across all calls

Find every call where a specific topic came up:

grep -l "payment" ~/ninja-talk/transcripts/*.txt

Count calls by date

ls ~/ninja-talk/transcripts/2026-05-25_*.txt | wc -l

Transcripts are plain UTF-8 text files. Use any tool you like: cat, grep, less, or download them via scp for offline review.

How your agent can use transcripts

Transcripts live on the same server as your OpenClaw agent. This means your agent can read them directly to learn from past conversations.

Review call quality

Ask your agent to read a transcript and critique the conversation. It can identify where the bot handled objections well, where it missed cues, and suggest playbook improvements.

Extract patterns

Have your agent scan multiple transcripts to find recurring questions, common objections, or frequent requests. Use these insights to refine your script profiles and FAQ responses.

Build training data

Transcripts are structured with clear speaker labels, making them easy to parse programmatically. Export them as training examples for prompt engineering or fine-tuning.

Version and availability

Automatic transcript saving was introduced in voice gateway v3.1.0. To check your version:

curl -s http://127.0.0.1:3335/ | jq .version

If you're on an older version, update your gateway by re-running the install command from your voice portal settings page. Your phone number, configuration, and subscription stay the same.

Storage

Transcript files are small (1-5 KB each). A deployment handling 100 calls/day produces roughly 10 MB/month. No automatic cleanup is applied. To manage disk space, archive or delete old transcripts as needed:

# Archive transcripts older than 90 days
find ~/ninja-talk/transcripts/ -name "*.txt" -mtime +90 -exec gzip {} \;

Quick reference