dev-util/kimi-cli-bin
Kimi CLI, Moonshot AI's terminal coding agent
Screenshots
View
Download
BrowseChangeLog
commit 2fc0b711cc0e90bb111aab68af2063e7bfa438bc
Author: lucascouts <lucascs@protonmail.com>
Date: Thu Sep 10 21:47:55 2026 -0300
add(dev-util/{opencode-bin-1.18.30, crush-bin-0.93.1, kimi-cli-bin-1.50.0}), add(licenses/FSL-1.1-MIT), add(metadata/md5-cache/dev-util/{opencode-bin-1.18.30, crush-bin-0.93.1, kimi-cli-bin-1.50.0}), mod(.autoupdate/{packages.toml, not-packageable.md})
Three more terminal agents, none of them in ::gentoo. opencode is the
most-starred open-source coding agent there is.
crush ships more than a binary: the tarball carries a manpage, bash/zsh/fish
completions and its licence, so all of it is installed. Its top-level
directory embeds both the version and upstream's own arch spelling
(crush_0.93.1_Linux_x86_64), which is why S is set in src_unpack per-arch
rather than assigned globally. doman also gets the page gunzipped first --
Portage compresses man pages itself and rejects one that arrives gzipped.
licenses/FSL-1.1-MIT is new. The Functional Source License is source-available
and converts to MIT two years after each release, so it is not OSI-approved
today and ::gentoo carries no FSL entry at all.
The kimi-cli autoupdate record uses a regex WITHOUT a leading "v", because
that tag really is a bare "1.50.0". Copying the v-prefixed pattern from a
neighbouring record would never match, and the probe would report "up to date"
forever while the package quietly went stale.
not-packageable.md gains an entry for the OpenHands CLI, and the reason is
worth stating because the name is now ambiguous three ways: the 87k-star
All-Hands-AI/OpenHands repo has become OpenHands/OpenHands and is TypeScript
(the Agent Canvas desktop app), the Python agent lives in
OpenHands/software-agent-sdk, and the actual terminal CLI is the PyPI package
"openhands". That CLI is small, MIT and has an sdist -- but resolving its
dependencies against ::gentoo, cut off at three levels and ignoring extras,
already leaves 65 packages missing, including litellm, browser-use, tokenizers
(Rust, its own CRATES block), fifteen tree-sitter-* grammars and the
macOS-only pyobjc. It is a volume problem, not a quality one, so the entry
records the evidence and the conditions that would reopen it rather than
pretending nobody looked.
Verified, not assumed: pkgcheck reports 0 findings on each of the three; each
completed a full `ebuild ... clean install` without root; and each installed
binary was executed -- opencode 1.18.30, "crush version v0.93.1", "kimi,
version 1.50.0". crush's image was checked to carry the manpage and all three
completion files.
Not verified: no merge onto the live system (no sudo on this host), so
directory ownership is untested.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Author: lucascouts <lucascs@protonmail.com>
Date: Thu Sep 10 21:47:55 2026 -0300
add(dev-util/{opencode-bin-1.18.30, crush-bin-0.93.1, kimi-cli-bin-1.50.0}), add(licenses/FSL-1.1-MIT), add(metadata/md5-cache/dev-util/{opencode-bin-1.18.30, crush-bin-0.93.1, kimi-cli-bin-1.50.0}), mod(.autoupdate/{packages.toml, not-packageable.md})
Three more terminal agents, none of them in ::gentoo. opencode is the
most-starred open-source coding agent there is.
crush ships more than a binary: the tarball carries a manpage, bash/zsh/fish
completions and its licence, so all of it is installed. Its top-level
directory embeds both the version and upstream's own arch spelling
(crush_0.93.1_Linux_x86_64), which is why S is set in src_unpack per-arch
rather than assigned globally. doman also gets the page gunzipped first --
Portage compresses man pages itself and rejects one that arrives gzipped.
licenses/FSL-1.1-MIT is new. The Functional Source License is source-available
and converts to MIT two years after each release, so it is not OSI-approved
today and ::gentoo carries no FSL entry at all.
The kimi-cli autoupdate record uses a regex WITHOUT a leading "v", because
that tag really is a bare "1.50.0". Copying the v-prefixed pattern from a
neighbouring record would never match, and the probe would report "up to date"
forever while the package quietly went stale.
not-packageable.md gains an entry for the OpenHands CLI, and the reason is
worth stating because the name is now ambiguous three ways: the 87k-star
All-Hands-AI/OpenHands repo has become OpenHands/OpenHands and is TypeScript
(the Agent Canvas desktop app), the Python agent lives in
OpenHands/software-agent-sdk, and the actual terminal CLI is the PyPI package
"openhands". That CLI is small, MIT and has an sdist -- but resolving its
dependencies against ::gentoo, cut off at three levels and ignoring extras,
already leaves 65 packages missing, including litellm, browser-use, tokenizers
(Rust, its own CRATES block), fifteen tree-sitter-* grammars and the
macOS-only pyobjc. It is a volume problem, not a quality one, so the entry
records the evidence and the conditions that would reopen it rather than
pretending nobody looked.
Verified, not assumed: pkgcheck reports 0 findings on each of the three; each
completed a full `ebuild ... clean install` without root; and each installed
binary was executed -- opencode 1.18.30, "crush version v0.93.1", "kimi,
version 1.50.0". crush's image was checked to carry the manpage and all three
completion files.
Not verified: no merge onto the live system (no sudo on this host), so
directory ownership is untested.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>