From 417 to 32 milliseconds: envsec on Effect 4 and Bun

The new beta (1.1.0-beta.1) wraps up a three-step journey: migrating from Effect 3 to Effect 4, trimming dependencies, and shipping a standalone binary built with bun build --compile. I measured every step on the same machine. The result: startup went from 417 ms to 32 ms (12.9× faster), and memory from 207 MB to 36 MB (5.8× less).

At a glance

The three steps

1. Effect 3 → Effect 4 (1.0.0-beta.18 → 1.0.2)

On Effect 3, a CLI like envsec pulled in a constellation of packages: effect, @effect/cli, @effect/platform, @effect/platform-node and, transitively, @effect/cluster, @effect/rpc, @effect/sql, @effect/workflow, @effect/printer, @effect/typeclass and more. 42 packages in total.

Effect 4 folds almost all of that into the effect package: the CLI module is now imported from effect/cli. The migration notes mention a fiber runtime rewritten for lower memory overhead and faster execution, and a core built for aggressive tree-shaking. For envsec, that meant:

• startup: −24% (417 → 318 ms)
• memory: −26% (207 → 154 MB)
• minified bundle: −28% (516 → 371 KB; gzipped 155 → 115 KB)
• packages: 42 → 19

None of the application logic changed. That gain comes from the library upgrade alone.

2. Leaner dependencies (1.0.2 → 1.1.0-beta.1 on Node)

Two targeted changes:

• Dropped @effect/platform-node. The CLI only needs NodeRuntime.runMain and NodeServices.layer, which that package just re-exports from @effect/platform-node-shared. Its barrel import, however, also loaded undici and the redis client on every start. envsec now imports the platform-node-shared modules directly.
• Replaced sql.js with node:sqlite. The metadata database used SQLite compiled to WebAssembly: a 23 MB package and a WASM module to instantiate on every start. It now uses the built-in node:sqlite module (Node ≥ 22.13, also supported by Bun). As a bonus, writes go straight to the file instead of re-exporting the whole in-memory database each time.

Result: startup 318 → 192 ms (−40%), memory 154 → 97 MB, packages 19 → 10, and the disk footprint nearly halved.

3. A standalone Bun binary

bun build src/main.ts --compile --minify --sourcemap --outfile envsec

--compile produces a single executable with the Bun runtime embedded: no Node.js, no node_modules, no module resolution at startup. The CLI code ships pre-bundled and minified in one file.

Result: startup 192 → 32 ms (5.9× faster), memory 97 → 36 MB.

The gain has two sources:

1. The runtime. On this machine an empty process takes 76.5 ms with Node (node -e 0) and 4.5 ms with Bun (bun -e 0). That alone is worth about 72 ms.
2. Loading the code. Subtracting the runtime cost, envsec on Node spends about 116 ms loading and initialising modules; the binary needs about 28. Reading one pre-bundled file is far cheaper than resolving and evaluating hundreds of files spread across 10 packages.

Where the 384 milliseconds went

Every step delivered a real gain, and each one enabled the next: without the move to node:sqlite the binary could not even be compiled, because the sql.js WASM file could not be located at run time.

Installation and distribution

The binary is not the lightest download: of its 61 MB, about 59 are the Bun runtime, so envsec and its dependencies weigh just over 1.5 MB. In exchange you don't need Node installed, there is no dependency tree to resolve, and startup is an order of magnitude faster. The npm package is still there for anyone who prefers Node, and it is 54% faster than the Effect 3 version too.

Methodology

• Machine: Apple M4 Pro, 24 GB RAM, macOS (Darwin 27.0)
• Runtimes: Node.js v26.10.0 for the npm versions; Bun 1.4.2 embedded in the binary
• Versions: installed from npm (envsec@1.0.0-beta.18, @1.0.2, @1.1.0-beta.1). The binary is envsec-darwin-arm64 from the v1.1.0-beta.1 GitHub release.
• Tools: hyperfine -N --warmup 5 --runs 40 for timings; /usr/bin/time -l (median of 15 runs) for memory; bun build --minify of the published dist/main.js for bundle size (sql.js excluded, since it loads as WASM).
• Data: an empty database isolated via ENVSEC_DB, so the measured commands never touch the system keychain.

Caveats

• This is one machine. Absolute numbers will differ on Linux and on Node 24 LTS: in earlier tests on Node 24, removing @effect/platform-node alone took --version from ~241 to ~149 ms.
• I measured startup, not long-running workloads. Commands that read secrets are dominated by keychain latency (Keychain, libsecret, Credential Manager), which is the same for every version.
• The binary doesn't use Bun's --bytecode yet, which could cut parsing time further.

Reproduce it

$export ENVSEC_DB=$PWD/db/store.sqlite && mkdir -p db
$hyperfine -N --warmup 5 --runs 40 \
$  -n "Effect 3" "effect3/node_modules/.bin/envsec list" \
$  -n "Effect 4" "effect4/node_modules/.bin/envsec list" \
$  -n "Bun"      "./envsec list"

Grab the binary from the v1.1.0-beta.1 release, or install it from the npm and Homebrew beta channels.

Conclusion

Effect 4 pulled its weight on its own: a quarter less startup time and memory just by upgrading the library. The real leap, though, came from three things together: a more compact Effect core, dependencies cut to the bone, and a runtime that starts in 4 ms. For a CLI you run all day, going from nearly half a second to 32 ms is the difference between a delay you notice and one you don't.