mirlyDownload
live ttft142ms·on apple silicon

Sound like
yourself.

A real-time copilot trained on your résumé, your STAR stories, your numbers. Sub-150ms first token. Invisible to screen share by documented OS API, not by trick.

free 7-min trial · no card · no auto-renew

mirly · listening
142ms
Q

Tell me about IAM permission sprawl at scale.

A
 
profile · sam patelclaude sonnet 4.6 · cached prefix

─── built on the best the field has shipped ───

Anthropic Claude Sonnet 4.6Deepgram Nova-3Apple ScreenCaptureKitwhisper.cppGoogle Gemini 2.5Groq LPUElectron 33sqlite-vecStripe billingVercel hostingAnthropic Claude Sonnet 4.6Deepgram Nova-3Apple ScreenCaptureKitwhisper.cppGoogle Gemini 2.5Groq LPUElectron 33sqlite-vecStripe billingVercel hosting

close-up · 01

It listens.
It thinks in your voice.
It whispers, in real time.

One floating window. Always on top. Invisible to screen share. The interviewer sees nothing — you see structured bullets in your own vocabulary, streaming faster than you can blink.

mirly · listening
142ms
Q

Tell me about IAM permission sprawl at scale.

A
• At Stripe the merchant team had 1,200+ IAM roles with overlapping policies.
• I built an IAM analyzer in Go that diffed real CloudTrail traffic against role policies.
• It opened minimal-policy PRs automatically — three rollout waves, four months.
• Cut 87% of unused permissions, ended at 312 roles.
• IAM-related Sev2 incidents went to zero for two consecutive quarters.
profile · sam patelclaude sonnet 4.6 · cached prefix
A

Question, transcribed live

Deepgram Nova-3 streaming partials at sub-150ms. Speech_final triggers the LLM only when the interviewer actually finishes — never on a fragment.

B

Answer, in your words

Claude Sonnet 4.6 with two cache breakpoints — base rules and your profile. After the first call, follow-ups are 5× cheaper and ~150ms faster.

C

Status, always honest

The header dot tells you whether stealth is verified for the platform you're on. If it ever turns amber, you'll know before you join the call.

close-up · 02

Same model. Same question.
Two voices.

Generic AI gives every candidate the same answer. Mirly rebuilds the system prompt every session from your résumé, your STAR stories, your target job — and is forbidden from contradicting what you actually shipped.

Without profilettft 811ms
• Audited existing IAM policies across AWS accounts and services.
• Implemented IAM Access Analyzer to monitor for unintended access.
• Leveraged AWS Organizations for centralized user and group management.
• Established a regular review cycle for IAM roles and policies.
• Documented baseline least-privilege policies using Config rules.
With profile · Sam Patelttft 854ms
• At Stripe the merchant team had 1,200+ IAM roles with overlapping policies.
• I built an analyzer in Go that diffed real CloudTrail traffic against role policies.
• It opened minimal-policy PRs automatically — three rollout waves, four months.
• Cut 87% of unused permissions, ended at 312 roles.
• IAM-related Sev2 incidents went to zero for two consecutive quarters.

live output, gemini 2.5 flash · captured in test-behavioral.mjs · replayed verbatim

specifications

Engineered to
a number.

01

First token

<0ms

p50, on-device whisper.cpp

02

Stealth

OS-API

NSWindow.sharingType + WDA_EXCLUDEFROMCAPTURE

03

Voice fingerprint

Per-user

résumé · STAR stories · target JD

04

Renewal

Never

opt-in renewal · 5-day reminder · zero auto-renew

benchmark

An order of magnitude
faster.

Multi-second lag is the #1 “AI tell.” A candidate who pauses 1.5 seconds before each answer reads as cheating. Sub-150ms reads as thinking on your feet.

Mirly
0ms
Parakeet AI
0ms
Cluely
0ms
LockedIn AI
0ms
Final Round AI
0ms

published or community-tested perceived first-token latency · mirly number from opt-in p50 telemetry

ready when you are

Seven minutes, free.
No card.

Pre-warmed connections. Cached prefixes. Zero auto-renew, ever.