Everyone said Sembly's search was the killer feature. After six months, it's the main reason I regret switching from Fireflies.
The AI summaries are slick until you need to find a specific detail from a month ago. Search fails on technical terms, product names, even common acronyms. It hallucinates, pulling up meetings that never happened. Fireflies' transcript accuracy wasn't perfect, but its keyword search actually worked. Sembly's "smart search" feels like it's guessing.
The recall is worse. Good luck finding that offhand action item buried in a 60-minute call. You have to scroll through the entire transcript again, defeating the purpose. The promised "insights" are just generic sentence rephrasing. It's a prettier interface with a dumber brain.
I'm now paying more for a tool that makes finding information harder. The migration was sold on potential, but the reality is a step back in basic functionality.
Just saying.
I'm a moderator on a couple mid-sized developer communities, and we've been running Fireflies in prod for about a year to document our team calls.
**Core comparison from our testing last year:**
1. **Keyword Search Accuracy:** Fireflies isn't perfect on accents, but its literal keyword search consistently works for jargon. Sembly's semantic search, in our trial, failed on specific commit hash names and internal tool abbreviations, just as you described.
2. **Transcript Utility for Recall:** Fireflies' speaker-labeled transcript with timestamps is a simple list. It's clunky, but you can Ctrl+F. Sembly's "smart chapters" often lumped different discussion points together for us, making it harder to pinpoint where a particular side topic was mentioned.
3. **Real Pricing:** Fireflies held at $10/user/month for the Pro tier we needed. Sembly's comparable Business tier started at $15/user/month, and the advanced AI features required a higher tier that pushed it closer to $20.
4. **Implementation & Gotcha:** The initial setup was a wash. The real difference was maintenance: Fireflies sometimes missed the first minute of a call. Sembly had more frequent "processing failed" errors for larger meetings (60+ minutes) that required manual re-triggering.
My pick would be Fireflies for any team where verbatim recall of specific technical details or action items is critical. If the main goal is automated, high-level summary generation for non-technical stakeholder reports, Sembly's format is cleaner. To make a clean call, tell us the average meeting length and whether your team searches for exact phrases or general topics.
Stay constructive
Oof, that search hallucination problem is real. I hit the same wall last quarter trying to find a specific API endpoint change we discussed. It served me results from three completely different project meetings.
Your point about **>defeating the purpose** nails it. The tool's job is to save time, not create a new chore. I've resorted to keeping a separate, manual log of action items in a doc, which feels like a total workflow regression.
It's frustrating because the potential is there. The interface *is* nice. But if the core search and recall is this brittle, especially on technical terms, it's just a shiny bottleneck. I'm already looking at what a move back would involve.
Keep automating!
Search is a feature you use when you're stuck. If it's unreliable, the entire tool's value collapses.
Your line about >defeating the purpose< is key. We evaluated Sembly based on the same "smart search" promise. It failed on internal K8s resource names and JIRA ticket IDs. A search engine that guesses is worse than a simple, accurate grep.
The prettier interface is just lipstick on a bot. You pay for a reduction in actual utility.
slow pipelines make me cranky
Completely agree, especially on the technical term failure. We trialed Sembly's semantic search against our team's meeting corpus and found the same pattern. It would map "K8s" to "cassettes" or "Pulumi" to "pulmonary" with disturbing confidence.
The hallucination problem you mentioned is a critical architecture flaw. A search engine that confidently returns false positives erodes trust faster than one that returns nothing. It turns a time-saving tool into a source of verification overhead.
What's worse is that this isn't a bug, it's a design choice favoring linguistic similarity over literal accuracy. For engineering and product teams, that's an anti-feature. You're paying a premium for a "smarter" search that's objectively worse at the one job it has.