By the numbers

A quantitative snapshot of the Zed codebase.

Data collected on 2026-04-30 at commit 24f62484e9. Numbers are derived from the working tree and git log.

Size

The codebase is overwhelmingly Rust.

xychart-beta horizontal
  title "Lines by language (top 8)"
  x-axis ["rust", "markdown", "json", "toml", "yaml", "python", "css", "javascript"]
  y-axis "Lines" 0 --> 1320809
  bar [1320809, 41514, 32140, 12069, 8622, 3257, 2258, 1020]

Language	Lines
Rust (.rs)	1,320,809
Markdown	41,514
JSON	32,140
TOML	12,069
YAML	8,622
Python	3,257
CSS	2,258
JavaScript	1,020

• Workspace crates: 231 (under crates/)
• Rust source files: 1,706
• First-party extensions: 6 directories under extensions/ (glsl, html, proto, test-extension, workflows, plus a top-level EXTRACTION.md/README.md)
• Helper scripts: 105 under script/

Largest crates

By Rust source lines (excluding generated/vendored files):

xychart-beta horizontal
  title "Top 10 crates by .rs LOC"
  x-axis ["editor", "agent", "project", "agent_ui", "gpui", "vim", "workspace", "collab", "ui", "debugger_ui"]
  y-axis "LOC" 0 --> 150000
  bar [147121, 100619, 90591, 63240, 62730, 46052, 44870, 42539, 26392, 25100]

Crate	LOC	What lives there
editor	147,121	Editor view, selections, decorations, completions
agent	100,619	AI agent core: threads, tools, history, prompts
project	90,591	Project model: worktrees, LSP, tasks, debugger
agent_ui	63,240	Agent panel UI
gpui	62,730	UI framework
vim	46,052	Modal-editing emulation
workspace	44,870	Panes, tabs, docks, the workspace shell
collab	42,539	Backend service for collaboration features
ui	26,392	Reusable design-system components
debugger_ui	25,100	Debugger panels and views

Activity

• Total commits: 37,439
• Commits in last 90 days: 2,783 (≈ 31/day)
• Repo first commit: Feb 20, 2021 (b400449a58)

Bot-attributed commits

460 of the 37,439 commits in history (~1.2%) carry a Co-authored-by: …[bot] trailer. This is a lower bound on AI-assisted work — inline AI tools (Copilot in-editor, IDE LLMs, the Zed agent itself when run interactively) leave no trace in git history, so most AI assistance is invisible here.

Tests vs. source

• Test files (matching *_test.rs, tests/, or #[test] heuristics): hundreds, distributed across crates rather than centralised.
• Many crates ship a test-support Cargo feature that builds shared test fixtures.
• Visual regression tests live under crates/zed/src/visual_test_runner.rs and snapshot images.

TODO/FIXME density

• Files containing TODO, FIXME, or HACK markers: 170 out of 1,706 Rust source files (~10%).
• See Cleanup opportunities → TODOs and FIXMEs is not a separate page in this wiki edition; grep crates/ for the markers if you are hunting cleanup work.

Reviewer pool

The REVIEWERS.conl file lists reviewers for ~20 areas (ai, gpui, lsp, vim, terminal, git, …) plus an <all> bucket of generalists. The "all" bucket has 10 reviewers, and the area buckets have 1–5 reviewers each.

Workflows

• .github/workflows/ contains 30+ GitHub Actions workflows: CI tests, deployment of collab, Cloudflare deploys, extension auto-bumping, release cuts, hotfix monitoring, contributor automation, and AI-assisted documentation suggestions.

Release channels

Three channels, configured in crates/release_channel:

• stable — what most users run.
• preview — opt-in early access.
• nightly — built each day from main.

script/cherry-pick, .github/workflows/cherry_pick.yml, and script/draft-release-notes automate the cut.

Caveats

• LOC is a crude metric and over-counts whitespace and macros.
• Git history numbers are sensitive to filter rules; git log --all may return different counts than git log on the default branch.
• "Bot-attributed" only counts trailers in commit messages, not authorship metadata.
• This wiki edition does not break out per-crate test-to-source ratios. If that is useful, it is straightforward to add via tokei or scc.