Hey everyone! I just had to share this little weekend project that’s already saving our marketing team so much time.
We’ve been using tl;dv to record and summarize all our customer feedback calls and internal syncs. The transcripts and AI notes are fantastic, but we kept hitting a snag: the “next steps” or action items identified in a call would just… live in the tl;dv summary. Someone had to manually copy them into our project management tool (we use ClickUp, but this would work for Asana, Jira, etc.).
So, I built a simple integration using Make (formerly Integromat). Now, whenever a recording is processed in tl;dv, the workflow grabs the video link, the summary, and specifically looks for a “Next Steps” section in the AI notes. It then formats that into a new task in our designated ClickUp list, assigned to the person who owned that part of the meeting agenda.
The setup was pretty straightforward. The key was using tl;dv’s webhook for “recording processed” and then a bit of parsing in Make to isolate the action items. If the AI summary doesn’t have a clear “Next Steps” header, it defaults to just posting the main summary so nothing gets lost.
The best part? It’s cut down our “meeting to action” lag from a day or two to literally minutes. Our project leads get pinged the moment the call wrap-up is ready. It feels like magic!
Has anyone else tried automating the handoff from tl;dv to other tools? I’d love to hear how you’re streamlining those workflows.
happy building
That's such a practical solution! I love how you used Make to tie those systems together.
Did you run into any issues with the AI notes formatting? Like, if the "Next Steps" aren't perfectly labeled, does your parsing still catch them? I'm thinking about doing something similar but worry about consistency.
Still learning.
> "if the 'Next Steps' aren't perfectly labeled, does your parsing still catch them?"
It doesn't. That's the first thing that broke when we tried the same thing. tl;dv's AI output is not consistent. Sometimes it's "Action Items", sometimes a bullet list with no header, sometimes buried in the summary paragraph. Regex on raw text is fragile.
We ended up calling the tl;dv API directly and grabbing the structured segments. They expose a "topics" or "sections" object in the webhook payload if you dig into the docs. That's more reliable than scraping the markdown summary.
If you're stuck with Make, pre-process the output with a regex filter that looks for common patterns. But prepare for a maintenance headache every time tl;dv updates their model.
Measure twice, migrate once.
Good approach, and the webhook is the right place to start. I've seen a few integrations like this break because they rely on the human-readable summary text, like the follow-up posts mention.
For long-term reliability, you should move your parsing logic to the structured data in the webhook payload, not the markdown. Look for the `sections` or `topics` array. It's more work upfront but saves you from tweaking regex every few months.
Also, consider adding a failover step that flags recordings where no structured action items are found, so they get a manual review instead of creating a useless task.