Five minutes after our sprint review ended, someone Slacked me: what did we decide about the deployment freeze?
I spent 45 minutes in that meeting. I had a transcript and a summary doc. I still had to scroll through 8000 words to answer a question I watched us debate live.
The problem was not the tools. It was how those tools thought about meetings.
The core insight
Every meeting tool treats every meeting as an isolated event. Transcribe it, summarize it, move on. But important conversations recur.
Sprint review 15 has context from sprint reviews 1 through 14. A sales discovery call sets up the demo. A quarterly 1:1 shapes everything that follows.
So we built around a different mental model: Series.
Instead of standalone meetings, you create a Series such as Dev Team Weekly, Customer Acme Corp, or Kim 1:1s. Every session builds on the ones before. By session 10, the AI tracks what changed, flags unresolved decisions from session 3, and surfaces patterns across your whole history.
We call this the Cumulative Context Engine.
What we shipped
- Real-time STT: Browser-based Web Speech API for low-latency feedback, plus server-side Voxtral (Mistral 4B) for 100+ language support
- AI flexibility: System-provided Gemini works out of the box, or bring your own Claude, GPT-4, or Gemini key for unlimited summaries (BYOK)
- Privacy first: Audio is not stored. Transcripts live in your workspace, encrypted at rest. No conversation data is used for AI training.
- Infrastructure: FastAPI + React, k3s production cluster, Cloudflare Tunnel, GitHub Actions CI/CD
What the community told us
Zoom/Meet bot requests
The single most requested feature. Our position: bot integrations require IT approval and add friction we deliberately avoided. O!TL;DR starts instantly - open app, start meeting, it records.
Zoom/Meet integration goes on the Q2 roadmap.
BYOK clarity
Developers immediately understood BYOK. Non-technical users found it opaque. Fix: a step-by-step visual BYOK onboarding flow, under 2 minutes, no API documentation required.
Free plan limits
The free tier gives 3 AI summaries per series. The irony: the Cumulative Context Engine requires more than 3 sessions to demonstrate itself. We are exploring a demo series with pre-seeded context so users can experience session 10 without waiting for it.
What this launch validated
Series as a mental model resonates. The recurring conversation framing clicked in a way meeting room never did.
Privacy-first positioning was welcomed. Especially from users in regulated industries: legal, healthcare, finance.
BYOK is genuinely compelling. Choosing between Claude, GPT, and Gemini was a consistent highlight.
What we are building next
- Search across meetings and series
- Simplified BYOK onboarding
- macOS desktop app (Tauri)
- Zoom/Meet integration research
- On-prem/self-hosted tier for regulated industries
Building in public means acting on what you hear. The top request goes on the roadmap. The clearest confusion becomes the next onboarding improvement.
Try it free at otldr.com. First summary in under 5 minutes.