Helm

Helm is a native Rust code editor for macOS. It aims to open in under 200 ms, edit TypeScript, JavaScript, Go and Rust with full language-server support, and host coding agents over the Agent Client Protocol. It has no Electron and no Node runtime.

What v1 must do is in docs/prd.md. How it gets built is in docs/plan.md.

Status

Phase 1 (headless core) is in progress. There is no window yet, so there is nothing to run as an app.

• T-101 Buffer trait and offset conversions — done
• T-102 Transactions with undo and redo — done
• T-103 Snapshots and concurrent readers — in progress
• T-104 Line endings and BOM preservation — done
• T-105 Selections and multiple cursors — in progress
• T-106 Command registry and keymap engine — not started
• T-107 VS Code bindings — not started
• T-108 Vim modes, motions, operators and counts — not started
• T-109 Vim text objects, registers, repeat and commands — not started

Crates

• helm — binary startup, settings file and crate wiring; currently a placeholder binary.
• helm-ui — GPUI editor views, tabs and splits, file tree, finder, terminal and agent panels. Empty.
• helm-editor — cursors, selections, editing commands and VS Code/Vim keymaps.
• helm-syntax — tree-sitter highlighting, folding, indentation and outline. Empty.
• helm-lsp — language-server client with lazy start, one server per root and idle shutdown. Empty.
• helm-project — files, watching, search, git and large files. Empty.
• helm-terminal — PTY and terminal emulator with capped scrollback. Empty.
• helm-agent — ACP client and edit transactions. Empty.
• helm-text — rope-backed text buffer, transactions, undo, snapshots and UTF-16 offsets.
• helm-bench — performance benchmarks; not shipped.

Building and testing

Rustup reads the pinned toolchain and installs rustfmt and clippy from rust-toolchain.toml. CI runs:

cargo fmt --all --check
cargo clippy --workspace --all-targets --locked -- -D warnings
cargo test --workspace --locked
cargo deny check --all-features --locked
cargo bench -p helm-bench

The PR bench job runs locked base and head benchmarks with --warm-up-time 3 --measurement-time 10, then compares them against the budgets. See AGENTS.md for the full local workflow and command notes.

How work happens

Each task starts with a spec PR containing stubs and ignored acceptance tests; after it merges, an implementation PR fills in the stubs and removes the test ignores. The protected-paths check requires the maintainer-only approved-protected label for protected changes, including edits to existing acceptance tests and benchmark budgets. The bench job compares each PR against its base commit on the same runner. See the task template and plan: How agents do the work.

Licensed under either of MIT or Apache-2.0, at your option.