pulse

one window for every local dev server. six tmux panes is still six tmux panes, and you know it.

brew install f4rkh4d/tap/pulse
# or
curl -fsSL https://pulse.frkhd.com/install.sh | sh
# or
cargo install --git https://github.com/f4rkh4d/pulse

cd your-project
pulse init    # scans for package.json / docker-compose / Procfile, drafts a pulse.toml
pulse

pulse reads pulse.toml, spawns every service, probes their health, watches their ports, and draws a tui with logs per service. restart one and its dependents bounce on their own. type t to tap live http traffic. type g for an ascii dep graph. that's most of it.

why another one of these

mprocs runs commands in panes. great tool, i used it for a year. pulse is what i kept wishing it did:

feature	pulse	mprocs	tmux	pm2
multiplex processes	yes	yes	yes	yes
per-service log panes	yes	yes	diy	yes
http health probe	yes	no	no	diy
port-in-use detection	yes	no	no	no
restart cascade on deps	yes	no	no	no
live traffic tap + replay	yes	no	no	no
ascii sentinel per svc	yes	no	no	no
config in one toml file	yes	yes	no	yes
auto-discover from repo	yes	no	no	no

pm2 is production-shaped and it shows in local dev. tmux is a multiplexer, not a process manager. mprocs is still the thing to beat.

what you get

• probes. per-service http poll, latency + rolling success rate in the sidebar
• port watch. [service.port] expect = 3000 → green :3000 bound badge, dim :3000 free when it's not
• traffic tap. tiny reverse-proxy records the last 500 requests per service. t opens the list, T splits on the latest request with headers + body preview
• dep restart. depends_on = ["postgres"] bounces web after postgres comes back. 1s grace, configurable
• ascii sentinels. one of five species per service (goblin, cat, ghost, robot, blob). they react to state and say things in the status bar
• auto-discover. pulse init reads package.json scripts, docker-compose.yml services, Procfile entries and drafts a config

sentinels

teaser, three species:

  ᨀ    goblin    "api back up. third time's the charm"
  ≋    blob      "postgres is just vibing at 5432"
  ᓚ    cat       "probe latency is weird, i'm watching it"

the robot species is pure ascii for fonts that don't like the others. pick per service via [service.agent] kind = "...".

gallery

auto-generate a config from an existing project. reads your package.json scripts + docker-compose.yml + Procfile and drafts a runnable pulse.toml:

see who's holding a port locally. no more lsof + grep dance:

stack examples

runnable pulse.toml for common stacks. copy into your project root, tweak cwds.

• next.js + postgres + redis + stripe
• django + postgres + celery + beat
• rails + postgres + sidekiq + js
• rust api + sqlite + nginx (docker) + esbuild

each sets probes on the right health endpoints, uses depends_on to cascade restarts, picks a different sentinel species, and sets up a traffic tap on the http service.

minimal config

[[service]]
name = "api"
cmd = "cargo run"
cwd = "./backend"
env = { PORT = "3000" }
color = "cyan"

[service.probe]
url = "http://localhost:3000/health"
interval = "5s"
timeout = "2s"

[service.port]
expect = 3000

[service.agent]
kind = "goblin"

[[service]]
name = "web"
cmd = "npm run dev"
cwd = "./frontend"
depends_on = ["api"]
color = "magenta"

save pulse.toml, run pulse. edit the file while it's running and the watcher diffs + restarts affected services.

keybinds

key	does
j / k	move up/down the service list (arrows work)
r	restart the highlighted service
s	stop the highlighted service
S	stop every service
enter	toggle logs focus (bigger view)
/	start a regex filter on the current logs
c	clear the current service's log buffer
t	open the tap panel for the selected service
T	split view on the latest tap request
g	full-screen dep graph
?	help modal with all keybinds
esc	close the current overlay
q / ctrl+c	quit. sigterm to children, sigkill if stuck

subcommands

• pulse init — scan cwd and draft a pulse.toml. reads package.json scripts (dev, start, watch, test:watch, serve), docker-compose.yml services, and Procfile entries
• pulse ports — list every process holding a LISTEN tcp port on this box. shells out to lsof, unix only
• pulse logs <service> [--lines N] — run a single service and print its logs. good for pipelines
• pulse share [--out path] — export current config as a single-file html snapshot
• pulse theme dump — print the default palette as a starter theme.toml
• pulse completions <shell> — emit a completion script. bash/zsh/fish/powershell/elvish

shell completion, one-liners:

pulse completions bash > /usr/local/etc/bash_completion.d/pulse
pulse completions zsh  > ~/.zsh/completions/_pulse
pulse completions fish > ~/.config/fish/completions/pulse.fish

benchmarks

measured on an m2 air, macos 14, cold cargo cache. ymmv, obviously.

thing	number
cold boot, 6 services, no probes	~180ms
cold boot, 6 services, http probes on	~210ms
resident memory, 6 services idle	~14 MB rss
per-probe bookkeeping overhead	~0.9 µs
tap proxy roundtrip overhead (loopback)	~0.4 ms p50

probe bookkeeping bench is in benches/probe_overhead.rs, run with cargo bench. boot-time numbers are stopwatch-grade, not statistical, so take them loosely. the tap overhead was measured with hey -n 5000 -c 20 against a local rust hello world with and without pulse in front.

compared to mprocs, again

short version: if you want something that works today, rock solid, big community, use mprocs. if you want probes + ports + tap + cascades in one config file, that's this. both are fine choices.

honest notes

• spike detection for the sentinels is a proxy: fast probe + recent stdout. a real req/s meter lands in 0.4
• the config watcher leaks a single Watcher box so it lives for the process. i know. still works
• sentinels use a couple of non-ascii glyphs (ᨀ, ʘ, ≋). on some fonts they render wide, on a few they become boxes. the robot species is pure ascii for those
• windows still untested. i don't have a windows machine and i'm not great at CI
• wrote this as a second-year project. it's held together with tests and optimism

license

MIT