[t3dotgg]

The background service runs t3 __service-launcher, a tiny supervisor that keeps the real server alive. It loaded the whole CLI just to reach its handler. In the packaged single executable, that is the full 18 MB module, so the launcher alone held ~260 MiB for as long as the service ran.

Now bin.ts is a 31 KB bootstrap. The service's exact t3 __service-launcher call runs straight from it. Every other command, including the launcher with flags or --help, loads the CLI from binMain.ts as before. In packaged builds, that CLI is a separate bundle embedded as a raw executable asset, so the executable is still one file.

Linux x64 packaged executables, medians of alternating runs against main:

	main	This PR
Launcher memory (PSS)	259 MiB	66 MiB (-75%)
Launcher time to start the server	467 ms	29 ms
Launcher CPU at startup	590 ms	20 ms
acp-mcp-call startup (wall / CPU)	288 / 294 ms	283 / 293 ms
t3 serve memory	420 MiB	418 MiB

Guards:

• build:exe fails if the bootstrap grows past 1 MiB, so the CLI cannot slip back into it.
• Both executable configs pass --require-module, so NODE_OPTIONS=--no-require-module cannot break the embedded load.
• The release smoke test now also runs __service-launcher on every platform.

Proof:

• check:changed, plus focused tests for the launcher, boot service, CLI loading, and the linux-arm64 executable. The CLI test fails if the bare-argv check or the launcher's full-CLI command is removed.
• Linux x64 packaged executable: version, launcher errors, launcher --help and flags, symlinked paths, --no-require-module, the speech worker over IPC, and SIGTERM stopping launcher and child.
• macOS arm64: built the real release archive on a Mac before the rebase, and the archive smoke test passed. Windows is covered by the release smoke test.

Council review

Rounds: 3

Delegate	Round 1	Round 2	Round 3
Claude Opus 5.5	7/10	8/10	9/10
Muse Wonder Full	7/10	9/10	9/10
Codex GPT-6.1 Sol	9/10	9/10	9/10

Created with Claude Opus 5.5 in Claude Code, running in T3 Code.

🤖 Generated with Claude Code

[macroscopeapp]

Approvability

Verdict: Needs Human Approval

Macroscope's review found this PR requires human approval — The change keeps the full CLI out of the service launcher's startup graph and packages it as an embedded asset. New diagnostic suppressions in bin.ts and vite.config.ts require human review.

No code changes detected at 9c7369f. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: 59ec1ba · PR result: 9c7369f · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: t3/explore-extreme-performance-wins
• File: .coderabbit.config.ts
• Commit: 9c7369f
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: 1a4725ec-4a63-41b1-b4b7-e738414c9c37

Commits

Reviewing files that changed from the base of the PR and between c7810c2 and 9c7369f.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

No actionable comments were generated in the recent review.

Walkthrough

The executable build now creates a separate bootstrap and CLI asset, validates bundle imports and bootstrap size, and embeds the CLI asset in SEA executables. Runtime startup loads the CLI through the bootstrap or runs the service launcher directly. Tests and an archive smoke check cover these paths.

Changes: CLI Bootstrap and Dispatch

Layer / File(s)	Summary
Build and embed CLI asset apps/server/vite.config.ts, apps/server/scripts/cli.ts, apps/server/scripts/cliErrors.ts, scripts/lib/linux-arm64-sea.ts	Executable packing creates a CLI bundle and a separate bootstrap. The build checks both bundles for external imports, rejects bootstraps larger than 1 MiB, and embeds the CLI asset in SEA executables.
Bootstrap loading and command dispatch apps/server/src/bin.ts, apps/server/src/binMain.ts, apps/server/src/cli/serviceLauncher.ts, apps/server/src/binCli.ts	The entry point loads the CLI from the embedded asset in SEA runs or imports it directly otherwise. It invokes the service launcher for the exact launcher command with no extra arguments. Other invocations use the CLI dispatcher. The service launcher awaits its work and sets the exit code on failure.
CLI startup and archive checks apps/server/src/cliCompileCache.test.ts, scripts/smoke-cli-archive.ts	Tests check CLI loading for normal and service-launcher invocations. The archive smoke check verifies the expected missing-T3CODE_HOME error.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Change: Refactor

Sequence Diagram(s)

sequenceDiagram
  participant BinEntry as bin.ts
  participant Launcher as runServiceLauncherCli
  participant CliMain as binMain.ts
  participant BinCli as binCli.ts
  alt Exact __service-launcher invocation
    BinEntry->>Launcher: run service launcher
    Launcher->>Launcher: await runServiceLauncher
  else Other invocation
    BinEntry->>CliMain: load CLI and call runCliCommand
    CliMain->>BinCli: import and run CLI
  end

Loading

Possibly related issues

• perf(server): every t3 CLI command re-parses the 13.5 MB bundle; Node's compile cache would cut t3 --help from ~550 ms to ~380 ms #19332: The changes separate bootstrap startup from loading the CLI main module, which relates to the issue’s focus on the cost of loading and compiling the full CLI bundle.

Suggested reviewers: maria-rcks, voltcrash

Merge Risk: Minimal · previous review c7810

This change moves the background service launcher into a small bootstrap to cut its memory use, and no merge-blocking defect was found. Windows and macOS SEA loading was not run directly and depends on the PR's reported release smoke tests.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title describes the launcher memory reduction, which is the main change.
Description check		The description covers the problem, change, and verification. It does not include the required scope and approval information.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Finishing Touches

Generate docstrings

• Commit to this branch
• Create a new PR

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: t3/explore-extreme-performance-wins
• File: .coderabbit.config.ts
• Commit: c7810c2
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: fbef11da-1bf9-40fe-90f5-ff75e54e7794

Commits

Reviewing files that changed from the base of the PR and between f56def2 and c7810c2.

Files selected for processing: 6

• apps/server/scripts/cli.ts
• apps/server/src/bin.ts
• apps/server/src/binCli.ts
• apps/server/src/cliCompileCache.test.ts
• apps/server/vite.config.ts
• scripts/lib/linux-arm64-sea.ts

Limit details: You’ve used all 10 included reviews currently available.

[Alvin-MN]

Independent test of this PR on Linux x64 (build:exe at head 61ccc10 against base 47f2c8d, Node 26.12.1, memory measured as PSS). Your numbers hold.

	base	this PR
Launcher, same binary file as the server	215 MB	31 MB
Launcher + server	581 MB	405 MB
Launcher, separate binary file from the server	248 MB	50 MB
Direct t3 serve	400 MB	383 MB

These also behaved correctly:

• SIGTERM with a live server: launcher exits 0 and the server stops.
• kill -9 on the server: launcher exits 1, the same as base.
• Missing home or state: same errors as base.
• Executable paths containing spaces, #, %20 and é, and a symlink.
• NODE_OPTIONS=--no-require-module.
• PR launcher running a base server.
• The PR merged onto main at a07be97: clean merge and build, launcher at 50 MB.

Two things that aren't in the PR description:

1. Existing installs may not get the saving. A launcher pinned by an older service unit stays large. Base launcher with PR server: 248 MB. A 0.0.46 launcher with PR server: 315 MB. Both run and stop cleanly, but the memory saving needs t3 service install or a fix for [Bug]: Service unit stays pinned to the runtime that installed it, so deleting old runtimes breaks the next restart #17490.
2. Small start-up cost on fast paths. About 36 ms (6%): acp-mcp-call went from 557 to 593 ms median, and acp-mcp-bridge from 557 to 589 ms (10 runs each, one batch per build). This looks like a fair trade, though acp-mcp-call pays it on every tool call.

Not tested: update handoff and rollback through the real update flow, Windows, macOS, and the web UI (client not built, so / returned 503).

I ran this with an AI coding agent.