“the suite takes six minutes.”
switch to slack. when it finishes you hear one sentence: pass or fail, the counts, and for a failure the one thing that broke. no stack trace read aloud.
“tell me when the tests finish”your agents ask. you answer out loud.
the only coding-agent notch app that listens and talks. claude code’s alert just says it’s waiting; swaey reads you the exact command and presses your answer in the right terminal. it never answers for you.
agent alerts are free on every plan. voice stays on your mac. prefer a key? ⌥↑ instead of the wake word.
never come back to “waiting for approval” again.
claude code’s own alert only says it needs permission. swaey reads the question off your terminal and tells you what it actually wants to run, with a warning first when it can’t be undone (rm -rf on a variable, a force push, a database reset). plans are read in plain words with the one risk to watch. multiple-choice questions come with their options, and codex prompts count too. answer from slack, figma or the kitchen: it switches to that terminal, presses your answer, and puts you back where you were. each project has its own chime, so you know which agent it is before a word.
yes
always
no
approve
skip step 3
the dark one
tell the api agent to add tests
what are my agents doing?
set it up once by saying “connect to my agents”. swaey adds a tiny hook to claude code that writes to a file only swaey reads.
swaey · just now
Claude Code in api wants to run: npm install zod. say yes, always, or no.> add input validation to the signup route
● Bash(npm install zod)
Bash command
npm install zod
Install the zod validation library
Do you want to proceed?
1. Yes 2. Yes, and don’t ask again for npm install in ~/code/api 3. No, and tell Claude what to do differently (esc)the interruptions, handled in a sentence.
the small things that break flow, dealt with while your hands stay on the code.
“port 3000 is already in use.”
swaey names the program, its command and which project folder it came from. “free it” stops it: only your own processes, only when you say so.
“what’s on port 3000? free it”“what was that docker command from tuesday?”
it searches your zsh, bash or fish history by what you remember, blanks any secrets, and copies the command. it never runs it.
“what was that docker command?”“a commit message that isn’t ‘fix stuff’.”
swaey reads the diff, matches your repo’s style (conventional commits if you use them) and types it into source control in vs code, cursor or windsurf. pr descriptions come with how-to-test steps. “commit it” refuses .env files and keys.
“write my commit” · “write the pr description”“what did i even do yesterday?”
your own commits across every repo since your last working day (friday’s, on a monday), plus uncommitted work and wip commits, as yesterday, today and blockers, copied for slack.
“give me my standup”“the agent touched 14 files. can i trust it?”
what changed, with a diagram, then file and line for what it read in the diff: weakened, skipped or deleted tests, silenced errors, fake or hardcoded data, look-alike helpers, secrets. it asks before running your tests and says the real result; “second opinion” gets a different model’s review. when an agent finishes while you’re away, it offers.
“can i trust what claude just did?”“claude keeps ignoring my claude.md.”
skills learned from the repo, its fix commits and your corrections in past sessions: verify, add-a-feature, stack gotchas and more, only what applies. hooks for rules that must always hold, a mistakes log whose repeats become gotchas, and “check my claude setup” for skills claude can’t see.
“make skills for my project”“a test i didn’t write is failing.”
swaey reads the failure off your terminal and types a prompt into your agent: quote the line that causes it, make the smallest fix, re-run to confirm. no swallowing the error, no skipping the test.
“fix this”“which of my agents is waiting on me?”
every claude code session at a glance: working, finished, or waiting on you. running codex, gemini, aider and cursor-agent sessions are counted too. give any idle one its next job without going to its window.
“tell the api agent to add tests”can’t type today? keep shipping.
if rsi, an injury or a disability makes the keyboard hard, you can still run your agents. say “hey swaey” instead of reaching for a key, hear what each agent wants and answer out loud, have swaey write the prompt, the commit message and the standup from what you say, and dictate into any app. hearing and speaking run on your mac.
what did i do yesterday? | your standup, copied for slack |
|---|---|
for claude code, add rate limiting to login | a proper prompt, typed into claude code |
yes· always· no | your answer pressed in the right terminal |
tell me when the tests finish | pass or fail, out loud |
write my commit | typed into the commit box in your style |
built for people with keys in their terminal.
a voice tool on a developer’s machine sees a lot. swaey is built to see as little as it can.
- your voice never leaves your mac. hearing and speaking run on-device.
- it looks at your screen only when you ask. no background screenshots.
- secrets are blanked before anything goes to the ai, shell history included.
- every request that left your mac is listed in the privacy tab.
- it never approves an agent’s request by itself.
questions developers ask
which agents does it work with?
claude code (permission questions, plans, multiple-choice questions and finished turns) and codex (its command approvals and questions), in terminal, iterm2, ghostty, warp, kitty, alacritty, hyper and wezterm. “what are my agents doing?” also counts gemini, aider and cursor-agent sessions that are running.
does it ever approve things on its own?
no. it tells you what the agent wants, warns you first when a command can’t be undone, and presses only what you said: yes, always, no, an option, or your own words for the agent. a plain “yes” to a plan keeps you approving each edit.
how does it know claude code is waiting?
say “connect to my agents” once. swaey adds a small hook to claude code’s settings that writes one line to a file only swaey reads whenever an agent needs you or finishes. sessions you already had open pick it up after a restart.
which editors does it type into?
vs code, cursor and windsurf for commit messages and prompts, and github’s pull request form. if it can’t confirm it’s in the right box, it copies instead of typing.
what does it cost?
agent alerts (the question read aloud and your yes, always or no) are free on every plan, forever, and don’t use your tries. everything else starts with 10 free tries, no card. pro is $15 a month or $120 a year (your local price at checkout). max ($100 a month) runs code checks, prompts and agent plans on our smartest ai, with about 7× pro’s usage. cancel anytime.
say it. keep coding.
your agents get answers. you keep your flow.
free10 tries, no card, no account
pro$15 a month for everyday use · cancel anytime
maxabout 7× pro’s usage, for all-day use
macOS 15 or later on apple silicon. prices shown in your currency at checkout.