Dictation that
sounds like you.

Hold fn, mumble the gist, paste a full reply. Telmur sees the thread, reads your tone, and writes the kind of message you'd write — if you had the time.

Free trial · 7 days, no card
Trusted by founders & engineers from
#eng-platform · Slack
# eng-platform · 12 members
MK
Maya Kowalski 2:04 PM
Heads up — staging migration is hitting timeouts on the auth shard. Should we slip the launch?
DL
Devon Liu 2:05 PM
+1, QA is still chasing edge cases on the new RBAC flow.
Message #eng-platform
Listening fn · Dictation
Whisper fn to stop
fn
01Press fn
02Speak
03Stream
04Pasted
/ The math

You speak 3.3× faster than you type.
That's a lot of gym time your coworkers won't know about.

Hours you spend typing per workday 3.5 h
0.5 h8 h
Your typing speed 45 wpm
hunt & pecktouch typist
9,450 words / day at 45 wpm typed.
Same words at 150 wpm spoken = 63 min.
147 minutes / day, back.
12.2h / week
across 5 workdays
76days / year
on a 250-day calendar
What you could do instead
rewatch all of Severance, twice.
Numbers are approximations. Your hands will thank you either way.
Built for what you already use
SlackiMessageVS CodeLinear NotionTelegramMailTerminal FigmaSafariCursorArc ObsidianSuperhumanChatGPTClaude GranolaRaycast
/ The trick

Mumble two words.
Paste a whole answer.

You're three messages deep in a Slack thread. You don't want to type a paragraph. So don't. Tell Telmur what you mean — it knows the rest.

What you said · 1.4s
"yeah, ok — let's slip to friday."
fn held · 1.4s 9 words
# eng-platform · Slack
MK
Maya Kowalski 2:04 PM
staging migration is hitting timeouts on the auth shard — should we slip the launch to next week?
DL
Devon Liu 2:05 PM
+1, QA still chasing edge cases on RBAC.
YO
You 2:06 PM · just now
yeah let's slip to Friday — not worth shipping over the RBAC stuff. Devon can you spin up a fresh staging shard before EOD so QA has the full week? will update the launch doc 👍
+ context: last 2 messages, channel topic, your last 50 sends 0.9s · pasted
🤫

Sounds exactly like you. Because it learned from you.

Telmur reads how you write in this channel — sentence length, the "yeah" instead of "yes", the lowercase "i", how often you use em-dashes, when you drop a 👍. The reply doesn't say "Telmur". It says you.

lowercase i em-dash habit "yeah" not "yes" 👍 at the end no exclamation marks
/ Four modes, one hotkey

Dictation is the default. The other three are when you want more.

Hold fn for dictation, or pick a mode with ⇧⌘19. Every prompt is an editable markdown file.

/ 01
Dictation.
The one you'll use 90% of the time.

Hold fn, talk, release. Telmur pastes the transcript — already capitalised, punctuated, and spelled with the names and jargon visible on your screen. No clean-up. Optional translation: speak Russian, paste English. ~99 languages.

fn · default Whisper + context 99 languages no edits needed
Before
// raw whisper output hey anastasia just looked at the figma i think the cta should be lime not amber slack me when youre back
After
Hey Anastasiia — just looked at the Figma, I think the CTA should be lime not amber. Slack me when you're back.
/ 02
Assistant.
When dictation isn't enough.

Sometimes you don't want to write the message yourself — you want an answer to what's on screen. Press ⇧⌘1, speak the gist, and Gemini composes the reply with the same screenshot for context. Auto-detects Slack, Telegram, WhatsApp.

⇧⌘1 screenshot + voice streams tone-aware
Before
// you say… "tell maya we're slipping to friday because of the rbac stuff"
After
Maya — we're going to slip to Friday. Want to make sure RBAC edge cases are buttoned up before we ship.
/ 03
Agent.
Multi-turn. Fresh eyes each turn.

A floating overlay for back-and-forth. Every turn grabs a new screenshot, so the AI keeps up with your IDE state, terminal output, or design draft. Streams in real time.

⇧⌘3 floating overlay live screenshots streaming
Before
$ npm run build ERR! ENOENT: no such file or directory, open './dist/server.js' $ _
After
You what's wrong here? Telmur Your build outputs to build/, not dist/. Either change start to node build/server.js, or set outDir: 'dist' in tsconfig.
/ 04
Rewrite.
Selection in. Selection out.

Highlight any text, press fn, dictate the change — "make it shorter", "fix the grammar", "translate to English" — Telmur replaces the selection in place. Works in any app that takes a paste.

⇧⌘4 in-place edit tone shifts translate
Before
I wanted to circle back and follow up on the action items we discussed in our previous sync to ensure alignment going forward.
After
Following up on what we agreed last sync.
FRONTMOST WINDOW Slack · #eng-platform 1280 × 800 · sent with request
FOCUS
/ Why the screenshot

It sees what you see —
so you don't have to clean up.

Telmur grabs the frontmost window with every request. Not to be clever — so the transcript matches reality. Names from the thread, jargon from the doc, tone from the channel. The clean-up step you usually do after dictation just… doesn't happen.

  • Names from the thread. If "Anastasiia" is on the avatar two messages up, that's how it's spelled.
  • Tone from the channel. Casual in #random, terse in #eng-platform, formal in a draft email.
  • Jargon from the screen. RBAC, ENOENT, Figma file names, your repo's branch names — picked up, not phonetic.
  • Frontmost window only. Never the whole desktop. Never persisted, never trained on. One click to disable.
/ For people who tune their tools

Everything is a file you can edit.

{ }

Markdown prompts.

Every system prompt is a plain markdown file. Tweak the assistant prompt, write your own. Changes pick up on next invocation — no restart.

~/Library/Application Support/
Telmur/prompts/
  assistant.md
  chat.md
  rewrite.md
  agent.md
Aa

Custom vocabulary.

A vocabulary.md of brand names, jargon, and people Whisper keeps mishearing. Telmur biases transcription toward your terms.

# vocabulary.md
- Telmur
- Kubernetes
- Anastasiia
- ENOENT

Hotkeys, mics, languages.

Pick the trigger (fn or ⇧⌘1–9), the input device (built-in, AirPods, or that USB mic), and the dictation target language. Persists per profile.

Trigger:   fn   (hold = PTT)
Mic:       AirPods Pro 3
Lang in:   auto
Lang out:  en
/ Pricing

One plan. Unlimited voice. No API keys.

All AI traffic is proxied through Telmur's backend. You don't bring keys. You don't manage credits. You just talk.

Telmur Pro

7-day free trial
$12/ month
  • Unlimited voice → text
  • Unlimited LLM requests (Gemini, proxied)
  • All four modes, all hotkeys, all languages
  • Editable prompts & vocabulary
  • Priority support, future model upgrades included
Start free trial
No credit card required. Cancel any time.
/ FAQ

Honest answers to the obvious questions.

Still wondering something? Email hi@telmur.com — a human reads it.

Where do my screenshots go?+
Each request sends a screenshot of the frontmost window over TLS to api.telmur.com, which proxies it to Gemini. We do not store the images, do not train on them, and do not retain logs of pixel content. Voice audio is sent to Whisper the same way and discarded after transcription.
Do I need an OpenAI or Gemini key?+
No. Telmur runs all traffic through its own backend with one token per user. Pricing is a flat subscription — no metering, no credits to top up, no bring-your-own-key gymnastics.
Which languages does dictation support?+
Whisper covers around 99 languages. Dictation auto-detects the input language; the optional translation step lets you choose any of those as the output. Mixed-language sentences usually work, too.
What about offline mode?+
Voice transcription and LLM responses both require the network — they hit our backend. Local-only Whisper is on the roadmap; the trade-off today is faster, larger models in the cloud.
Will it work with my mic / AirPods / USB interface?+
Yes. Telmur lets you pick any input device CoreAudio sees, and remembers it per profile. We've tested with AirPods Pro 3, Shure MV7, the built-in mic, and most USB interfaces.
Is the app signed and notarized?+
Yes. It's a Developer-ID-signed, notarized native macOS app, around 43 MB. No Electron, no helper processes, no kernel extensions. macOS 13 (Ventura) or later, Apple Silicon or Intel.
Can I cancel?+
Whenever you want, from the in-app menu or via email. Cancellation takes effect at the end of the current billing period. The free trial is seven days, and we don't ask for a credit card up front.
/ Download

Stop typing what you could just say.

GDPR-compliant. Independently audited (SOC 2 Type II). Your audio is never used for training — ever. Seven days free, no credit card.

Download Telmur.dmg Read the docs
Try it the second it opens — fn hold & speak