What build mode does it pick for Python-only edits under an editable install?

None. It detects that only Python files changed under an editable install and skips the build.

What does it do for C++ or Cython changes?

It runs targeted Bazel builds for the affected modules instead of a full rebuild.

When does it run a full source build?

On first setup, when no prior build exists.

Can it build a single module like raylet or GCS?

Yes. It includes commands for building individual modules such as raylet and GCS.

Are there options for debug builds or low-memory machines?

Yes. It offers debug and fast build options, plus options for low-memory machines.

Rebuild Ray

SkillDev tools

The rebuild skill guides an AI agent through rebuilding Ray from source. It checks which files have changed with git diff and picks the matching build path, so the agent runs the right build mode instead of a full rebuild every time. This covers Python-only edits under an editable install, targeted Bazel builds for C++ or Cython changes, a dashboard build, or a full source build on first setup.

Use Rebuild Ray in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add Rebuild Ray and connect your AI. About a minute.

Claude ChatGPT Ahel Desktop

Also: Claude Code · Cursor · Codex

Then ask your AI: use the Rebuild Ray skill

Details

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Account requirements not reviewed. Check the skill instructions before use; ahel provides instructions and does not run this skill.

No other account needed.

Have a Ray source checkout with git available so changed files can be detected.

Rebuild RayStart free

What your AI can do with it

Check changed files with git diff to choose a build mode
Skip building when only Python files changed under an editable install
Run targeted Bazel builds for C++ and Cython changes
Build the dashboard when dashboard files change
Run a full source build on first setup
Build individual modules such as raylet and GCS

Getting started

Have a Ray source checkout with git available so changed files can be detected.
Have the build dependencies Ray requires for source builds installed.
Add the rebuild skill to your agent so it can run build commands.
Ask the agent to rebuild Ray after making changes, and it will pick the build mode from the diff.

What this skill tells your AI

The instructions your AI receives, as published by ray-project/ray in .claude/skills/rebuild/SKILL.md and read by ahel’s review.

Canonical build docs: doc/source/ray-contribute/development.md Use the user's configured Python from CLAUDE.local.md, or fall back to which python. Update this skill if any changes are detected in development.rst.

Determine what to rebuild

Check what files changed: git diff --name-only HEAD (or ask the user)
Pick the appropriate build mode:
- Only .py files changed (no .h, .cc, .pyx, .pxd) → Python-only rebuild (not required if installed with pip install -e . — editable mode auto-picks up Python changes)
- .h, .cc, .pyx, .pxd files changed → C++/Cython rebuild
- Dashboard files changed (python/ray/dashboard/client/) → Dashboard rebuild first
- First time / clean build → Full source build

Running Python tests that exercise C++ changes: When a Python test depends on C++ or Cython code (raylet, GCS, core_worker, etc.), run the C++/Cython rebuild below before running the test. Editable Python installs do not rebuild the compiled extensions, so stale binaries will run otherwise.

Python-only rebuild

For changes to Tune, RLlib, Autoscaler, and most Python files. Two approaches depending on how Ray was installed:

If installed with `pip install -e .` (editable/full source build)

No rebuild needed — Python changes are picked up automatically.

If installed from nightly wheel + setup-dev.py

Ensure a nightly wheel is installed (see doc/source/ray-contribute/development.md for the correct wheel URL for the user's platform and Python version)
Link local files: python python/ray/setup-dev.py
Can skip specific dirs: python python/ray/setup-dev.py -y --skip _private dashboard

Warning: With setup-dev.py, don't run pip uninstall ray or pip install -U. Instead: rm -rf <site-packages>/ray, reinstall wheel, re-run setup-dev.py.

C++/Cython rebuild (after initial full build)

Full package rebuild

bazel run //:gen_ray_pkg

Build individual modules (faster)

bazel build //src/ray/raylet:raylet          # Raylet only
bazel build //src/ray/gcs:gcs_server         # GCS server only
bazel build //src/ray/core_worker:core_worker # Core worker only
bazel build //:ray_pkg                       # All C++ and Cython

Build variants

bazel run -c fastbuild //:gen_ray_pkg — fast build (less optimization)
bazel run -c dbg //:gen_ray_pkg — debug build with symbols (for gdb)
bazel run -c opt //:gen_ray_pkg — optimized build

Make permanent in ~/.bazelrc: build --compilation_mode=fastbuild

Full source build (first time)

If dashboard changed: cd python/ray/dashboard/client && npm ci && npm run build && cd -
Install deps: cd python/ && pip install -r requirements.txt
Build: pip install -e . --verbose

Resource-constrained builds

Add to ~/.bazelrc if machine runs out of memory:

build --local_ram_resources=HOST_RAM*.5 --local_cpu_resources=4
build --disk_cache=~/bazel-cache

Worktree notes

Verify build paths: bazel info execution_root
If build fails, check .bazelversion matches installed Bazel
Each worktree may need its own pip install -e .
If in a worktree and bazel targets fail, ensure symlinks are correct

Signals

GitHub stars: 44k
Forks: 8k
Last commit: Oct 2026

ahel review

K1binfo
installs-packages

Automated review, not a security audit. Ruleset v1+k2.

Questions

What build mode does it pick for Python-only edits under an editable install?: None. It detects that only Python files changed under an editable install and skips the build.
What does it do for C++ or Cython changes?: It runs targeted Bazel builds for the affected modules instead of a full rebuild.
When does it run a full source build?: On first setup, when no prior build exists.
Can it build a single module like raylet or GCS?: Yes. It includes commands for building individual modules such as raylet and GCS.
Are there options for debug builds or low-memory machines?: Yes. It offers debug and fast build options, plus options for low-memory machines.

Advanced

Item type: skill
Key: rebuild
Source: github.com/ray-project/ray

github.com/ray-project/ray