Your AI pair can read
faster than you can type
Modern development shifted the bottleneck: less time writing code, more time describing what you want — to Claude, ChatGPT, Copilot, an agent in a terminal. That's a writing job, and typing caps it at ~40 words a minute while you think at ~150 (the research). Dictation removes the cap.
Where voice fits a developer's day
Prompting and brain-dumps
The richer the context you give an AI assistant, the better the output — and rich context is exactly what's tedious to type. Speak the whole architecture ramble: constraints, edge cases, what you tried, what you suspect. Blabb cleans the "ums" and false starts; the model gets the full picture. Long prompts stop costing anything.
Everything that isn't code
Commit messages, PR descriptions, code review comments, standup notes, tickets, the Slack reply explaining why the deploy is delayed. All prose, all day, all dictatable. Six writing styles mean the ticket can be terse and the customer reply friendly — same voice, one hotkey.
Terminals and remote sessions too
Blabb types key-by-key into whatever has focus — terminals, VS Code, SSH sessions, even clipboard-blocked RDP/Citrix sessions where nothing else lands (how). One Windows limit applies to all tools: no typing into elevated/admin windows (UIPI).
What voice is bad at (and what fixes it)
Don't dictate syntax
Speaking code token-by-token ("open paren, lambda x") is slower than typing it. The winning split: dictate intent and prose, type (or let the AI write) the code. If you need full voice coding for accessibility, purpose-built tools like Talon serve that; Blabb is for the words around the code.
Jargon needs learning
Generic recognizers butcher identifiers — "kubectl", "PostgreSQL", your service names. Blabb's glossary learns them from your corrections and applies them next time, locally. A week in, your vocabulary is part of the tool.
Privacy of proprietary context
Prompts routinely contain proprietary architecture and unreleased plans. Cloud dictation adds a second company to every prompt's audience before it even reaches your AI provider. Blabb processes speech on-device — the only recipient of your words is the app you're typing into.
SHAs, IDs, keys, IPs — kept verbatim
A commit SHA in a PR description, a ticket ID, a server IP, a license key: the cleanup step that polishes your prose would quietly mangle them. Blabb holds those tokens aside and restores them exactly as dictated — proven against long digit runs and MD5/SHA hashes — so the polished sentence still pastes where it has to.
Brain-dump at 150 wpm. Let the model argue back.
Two weeks free. Speech and cleanup both run on-device.
FREE TIER 2,000 WORDS/DAY · 14-DAY UNLIMITED TRIAL IN-APP · CANCEL ANY TIME