When Good Ideas Strike
Ideas, the good kind, always strike when our hands are full
Welcome
This is the first issue of the ‘Roam 'n' Around Newsletter, where we take a look at how RoamResearch users are solving problems inside their graphs.
This issue dives into a specific yet familiar challenge for many Roam and knowledge workers who are juggling work projects, personal projects while also raising children.
Problem
You’re mid-diaper change, or lulling your baby to sleep, when bam a breakthrough thought hits. But your hands are busy, and Roam is somewhere on a different screen, behind a passcode, across the room.
What to do before the idea vanishes?
Solution
JimmyLV has found a workflow that’s as smooth as it is clever.
Voice as the bridge, AI as the sorter
Here’s how they do it:
Voice Input with Live AI Extension: Jimmy uses the Live AI extension in Roam, paired with the BibiGPT API, to record voice notes. Unlike standard voice recognition tools like Whisper, BibiGPT excels at handling non-English languages (e.g., Chinese), ensuring accurate transcription.
Setup: Jimmy configures multiple API keys, routing OpenAI requests through a proxy to BibiGPT’s base URL for enhanced voice recognition.
Workflow: While multitasking, Jimmy speaks their ideas, and Live AI/BibiGPT transcribes them into Roam Research. Live AI then cleans up and structures the brain dump as summaries, actionable to-dos or structured blocks with relevant tags.
Organizing with Chat Roles: While not always used, this allows for tagged context when AI replies are inserted—so the “author” of each AI block has full Roam page powers. Jimmy uses LiveAI’s feature to attribute AI-generated responses to specific AI model names which integrates with Roam’s page capabilities.
Outcome: Jimmy’s graph captures raw ideas instantly, preserving context without interrupting their day. These notes are ready for deeper processing later
The Tools:
RoamResearch
Live AI Extension (Roam Depot)
BibiGPT proxy setup
Custom LLM API keys
Chat Roles (optional but useful)
Thoughts
Why it Clicks:
The genius here is delayed structure. Rather than trying to speak in perfect blocks, he leans into the mess, then lets Roam + AI make sense of it. Leaning into the mess has actually been Roam’s promise from the beginning. The separation of input and processing allows for fluid thinking, while Roam + LiveAI’s flexibility absorbs the outcome.
Opinion:
I have no idea why Fabrice Gallet, primarily designed LiveAI as an AI voice capture tool. But he was definitely on to something. As AI continues to be an entrenched part of the fabric of society, I think it is going to be increasingly normal for people to be speaking to their phones/computers. Anecdotally, whenever I am hanging out with my 7–10 year-old nieces and I hand over my phone for them to play with chatGPT. Its instinctive that they go straight to the audio button and start speaking to it. I didn’t show them that, it’s just what they do. It’s how they text also.
Your Turn
Have you tried using voice input in Roam? What worked? If not, is this something worth giving a try?
next issue: We dive deeper into JimmyLV’s graph and take a look at how he’s building his own app inside Roam.