What meeting capture tool actually works for hybrid work in a mid-market firm?

Alright, let's set the scene. My team is fully distributed across three time zones, but we still have these critical, high-stakes meetings with external partners who often prefer a hybrid setup—some in a conference room, some dialed in. The audio soup is real. We've been on a quest for a meeting capture tool that doesn't just record, but actually *integrates* the output into our workflow. We tried a few, and Fireflies.ai keeps coming up.

I'm curious about experiences from a mid-market infrastructure or platform engineering perspective. We're not a tiny startup, nor a massive enterprise. Our needs are specific:

* **Multi-source audio capture:** Can it reliably distinguish and transcribe voices from a room mic, a Polycom, and individual Zoom/GMeet attendees without creating a garbled mess?
* **Action Item Extraction:** This is huge. We live in Jira and GitHub Issues. Does the AI actually identify actionable *technical* tasks? ("We need to increase the HPA threshold on the payment service" vs. generic "follow up").
* **Integration Depth:** Slack notifications are table stakes. I'm talking about **GitOps-friendly outputs**. Can I get a structured summary (maybe as a markdown file or even a JSON blob) pushed to a repo, or an automated ticket creation via webhook with specific labels?
* **Security & Compliance:** Where is the data processed? Can we enforce retention policies? If it's generating summaries, where are those stored?

We did a proof-of-concept with another tool that shall not be named, and the YAML it generated for a "summary webhook" was... disappointing. It was just a basic POST. I'm dreaming of something that could trigger a simple GitHub Action to file an issue.

```yaml
# Example of an ideal webhook payload structure for our use case
actionItems:
- owner: "platform-team"
text: "Review Istio telemetry config for api-gateway namespace"
source: "00:12:34"
confidence: 0.89
- owner: "devops"
text: "Scale up node pool for preprod cluster before load test"
source: "00:23:15"
confidence: 0.94
meetingMetadata:
kubernetesTopics: ["HPA", "Istio", "node pools"]
participants: ["alice@example.com", "bob@partner.com"]
```

So, for those of you in similar hybrid, mid-market tech trenches:
1. Does Fireflies.ai handle the chaotic hybrid audio reliably?
2. How smart is the action item detection for technical workflows?
3. Have you managed to wire it into a CI/CD or GitOps pipeline? Show me your configs or Helm values! 😄
4. Any deal-breaking pitfalls on the security or pricing side as you scale?

I'm ready to dive into the details. The goal is less "meeting transcription" and more "meeting-derived artifact automation."