You don't have an AI problem. You have a detour problem.
Most people who "use AI all day" are really doing something more tedious than that: they're commuting to it. You're deep in a function, or three sentences into a hard email, and you hit a question. So you stop. You Alt-Tab to a browser tab or a chat app. You retype the question, or paste in the paragraph the model needs as context. You wait. You read. You copy the answer. You Alt-Tab back to where you actually were. You paste. You find your place again.
That round trip is the real cost of how most of us use AI, and it has a name: a context switch. The model is fast. The detour is not. And you pay it dozens of times a day, every time the smartest tool on your machine lives behind a separate window you have to go visit.
davr removes the trip. You press a hotkey, speak your question out loud, and Claude's answer is typed in — right where your cursor already is, in whatever app you're already in. No tab. No paste. No copy-back. The AI comes to you.
The friction nobody budgets for
The individual detour feels cheap. Two seconds to switch, a few to paste, a few to bring the answer home. It's the frequency that gets you.
A 2022 Harvard Business Review analysis found that the average digital worker toggles between applications and websites nearly 1,200 times a day. Separately, the often-cited University of California, Irvine research on interruptions puts the cost of a full context switch at around 23 minutes to fully refocus on the original task. You won't pay the full 23 minutes for a quick AI question — but you will pay something every single time, and the something compounds. The tab you opened to ask one thing is still open an hour later with four more things in it. The flow state you had while coding is gone the moment you leave the editor.
The point isn't the exact number. The point is the shape of the problem. When the AI is a place you go, using it is always an interruption of the thing you were doing. The interruption is the product's fault, not yours.
The davr workflow: ask in place
Here's the whole loop with davr:
- You're working in some app — an editor, an email, a doc, a terminal.
- You press the davr hotkey and speak your question.
- Claude's answer is typed in, wherever your cursor is.
That's it. You never left the app. The context didn't have to be pasted anywhere, because you never went anywhere. The answer arrives in the document you're building, not in a chat log you'll have to ferry it out of.
This is a small change that feels like a large one, because it removes the part of "using AI" that was never about thinking — the logistics. Asking becomes a keystroke instead of an errand.
What it actually looks like
A few concrete moments where asking-in-place changes the texture of the work:
Coding in your IDE
Your cursor is in the editor and you hit a wall: the exact regex for a US ZIP+4, the flag you always forget on a CLI command, the one-liner to flatten a nested array. Hotkey, ask, and the answer is typed inline — you keep your hands and your attention in the editor. For the heavier version of this — dictating whole blocks of code by voice — davr has Vibe Coding on the Pro tier. But for the quick "how do I…" interruptions, asking in place is the unlock.
Drafting an email
You're three sentences into a message in Outlook or Gmail and you need a phrasing you can't quite find — a polite way to push a meeting, a softer version of a hard "no." Hotkey, ask, and the suggested line drops into the body where you're typing. You edit it to sound like you and keep going. The message never had to leave the compose window.
A quick fact or summary
Not every question deserves a research session. Sometimes you just need a definition, a conversion, or a two-line summary of a concept so you can keep writing. Asking in place is the difference between a five-second answer and a five-minute side quest that ends with you reading three tabs you didn't mean to open.
Rewriting what's already there: Transform Text
Asking isn't only about generating something new. Often the text already exists and just needs to change. With Transform Text (Pro), you highlight a sentence or a paragraph, speak an instruction — "make this more formal," "tighten this to two sentences," "rewrite this in Spanish" — and the selected text is rewritten in place. Same principle as Ask Claude, pointed at words you've already written instead of a blank space.
And when the job is bigger than a rewrite — turning a spoken brain-dump into a finished blog post, a social thread, or a polished email — that's Briefs, davr's voice-to-finished-content pipeline on the Max tier. Different scale, same idea: your voice drives the work, in the place you're already working.
The bigger idea: an ambient command layer, not a chatbot
Step back and the pattern is clear. There are two ways to put a powerful AI in front of a person.
You can build a destination — a chat window the user visits, pastes context into, and copies results out of. That's the model almost everyone ships, and it's why "using AI" so often means tab-juggling. The intelligence is real, but it's quarantined in one window, and getting your actual work in and out of that window is on you.
Or you can build a command layer — make the AI available everywhere, on a keystroke, so it acts on the words and the app already in front of you. The AI stops being a place and becomes a capability of your whole desktop, the way copy-paste or spell-check are. You don't "go to" spell-check. It's just there, in every app, when you need it. That's the bar.
This is also where davr's broader thesis shows up. Plain dictation — turning your voice into text in a box — is table stakes now; plenty of tools do it well. The interesting question is what happens to the words after they're text. Do they just sit in a field, or do they drive an action: ask a question, rewrite a paragraph, translate, run a command? Ask Claude from a hotkey is one of the clearest examples of the action layer being the point, not the dictation. (More on that idea in [link: dictation-is-the-input-action-is-the-point].)
It's worth being precise and fair about where other tools sit here, because the line is subtle. Wispr Flow — a genuinely fast, polished dictation app — has a Command Mode that does rewrite selected text in place, which is good work. But by its own documentation, when you ask a question without selecting text, Command Mode opens Perplexity in your browser with your query . That's the tell. Even a strong dictation tool's instinct, when you ask it something, is to send you to a browser — to a destination. davr's instinct is to bring the answer to the app your cursor is in. Both are reasonable designs; they're just aimed at different problems. Dictation tools are very good at getting your words into an AI's box. davr is built to bring the AI to you. (We go deeper on the full comparison in [link: davr-vs-wispr-flow].)
Whose key, whose account: the privacy note
Ask Claude is an LLM call — your question goes to Claude, because Claude is what answers it. There's no on-device version of that; the model lives in the cloud. So the honest privacy question isn't "does anything leave the machine" (it does — that's how you get an answer). It's whose account the request runs through.
davr runs on bring-your-own-key (BYOK). You connect your own OpenAI and Anthropic API keys, stored in Windows Credential Manager rather than sitting in plaintext somewhere. When you ask Claude, the request goes to your Anthropic account — not through a davr middleman that gets a copy along the way. You're billed cents by your provider for what you actually use, and the data path is one you already control and can audit. BYOK is the Free tier at $0, and it stays the cheaper option on Pro ($6/mo) and Max ($10/mo) as well — your keys, uncapped, private.
Dictation is a separate story. For the transcription step, davr offers a Local option (on-device Whisper, audio never goes to OpenAI) and a Privacy Mode toggle that turns off the Claude cleanup pass (so that text never goes to Anthropic). Flip both on and dictation is fully on-device. Ask Claude sits outside that envelope by design — it's a deliberate network call to a model, but a call to your account, on your own terms, not an opaque hop through someone else's server.
Who this is for
If you keep a chat tab pinned in a second window and Alt-Tab to it twenty times an hour, this is for you. Developers who break flow to look something up. Writers and PMs who draft in one app and "ask the AI" in another. Founders doing six jobs who need a fast answer typed into the thing they're already in, not a new tab to babysit. Anyone whose real complaint about AI isn't that it's not smart enough — it's that reaching it is a chore.
Try it
davr is free with your own API key — connect your OpenAI and Anthropic keys, pay your provider cents a day, and ask Claude from a hotkey in any app at $0 to davr. If you'd rather not wire up keys yet, there's a 14-day, no-card trial of the managed AI features (about 2,500 words a week) so you can feel the in-place workflow before deciding anything.
davr runs on Windows today. Mac, iPhone, and Android are rolling out over the next couple of months.
Press a hotkey. Ask your question. Keep working in the same window you were already in. That's the whole idea — and once asking is a keystroke instead of a trip, you stop visiting your AI and start just using it.