Skip to content

release: 0.4.20 #177

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
asher merged 1 commit into from
Oct 4, 2026
Merged

release: 0.4.20 #177

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
101 changes: 52 additions & 49 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,8 @@ adhere to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [Unreleased]

## [0.4.20] - 2026-10-04

### Added

- `gmlx launch <name>` runs a custom agent of your own from `launch.agents`
Expand Down Expand Up @@ -104,55 +106,6 @@ adhere to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
the menu bar's Edit config does not save, and the message names the file
that the link leads to.

### Removed

- A bare `gmlx serve` with no config no longer serves the GGUF files of the
current folder, and says to run `gmlx init` instead. A login item or menu
bar autostart set up for such a server fails until a config exists, and
`gmlx doctor` warns about one.
- `gmlx init` no longer offers to write the config into the current folder.
- The server no longer serves mlx-vlm's `/v1/realtime` WebSocket route,
which skipped the API key check.
- The server no longer serves mlx-vlm's `/v1/settings` route, which could
point the prompt cache and the drafter at any folder, or its own
transcription routes, which decoded an upload with ffmpeg before any
check.

### Security

- A request's `chat_template_kwargs` must be an object, and can no longer
replace the model's chat template or set another parameter of the
template call. Such a request gets a 400, so the server never renders
Jinja that a client sends.
- A server, menu bar or extra install that gmlx starts no longer imports a
`gmlx` or `pip` package from the current folder in place of the installed
one. The processes gmlx starts also drop the empty and relative entries of
`PYTHONPATH`, which put the current folder on the import path.
- `gmlx launch` creates a configuration file with mode 600 and keeps the
mode of each file it rewrites, so a file that holds a key stays private.
It writes through a symbolic link that stays inside your home folder
instead of replacing it, and refuses a link that leads outside.
- A server on a loopback address refuses a request whose address it cannot
read, and one that reaches it from another address through a redirect,
such as the one a localhost domain of Apple container adds. It also no
longer takes a client's address from `X-Forwarded-For`.
- `gmlx launch open-webui` runs Open WebUI on `127.0.0.1` only and sets its
`CORS_ALLOW_ORIGIN` to its own address, so other computers cannot reach it
and other web pages cannot read its answers. Pass `-- --host 0.0.0.0` to
listen on every address, and export `CORS_ALLOW_ORIGIN` for another
address, such as that of a reverse proxy.
- `gmlx init`, `gmlx pull`, `gmlx sync-models` and `gmlx rm` keep the mode
of the config file, and `gmlx init` gives a new config mode 600. These
commands and the menu bar's Edit config write into the file that a config
link leads to, instead of replacing the link with a plain file.
- When you press Tab, bash completion no longer runs a command, such as
`$(...)`, that a model id, alias or assistant name in the config holds.
Completion leaves out such a name, so a script from 0.4.19 that you saved
to a file is safe too.
- The menu bar, `gmlx service` and the server run `launchctl`, `ps`,
`sysctl`, `vm_stat`, `codesign`, `open`, `pgrep` and `pbcopy` by their
full paths, never through `PATH`.

### Fixed

- `--kv-bits` on gpt-oss keeps an fp16 cache and says why, where `run`
Expand Down Expand Up @@ -273,6 +226,56 @@ adhere to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
such as `npx` and `node`, in `/opt/homebrew/bin`, which the `PATH` of the
login item leaves out.

### Removed

- A bare `gmlx serve` with no config no longer serves the GGUF files of the
current folder, and says to run `gmlx init` instead. A login item or menu
bar autostart set up for such a server fails until a config exists, and
`gmlx doctor` warns about one.
- `gmlx init` no longer offers to write the config into the current folder.
- The server no longer serves mlx-vlm's `/v1/realtime` WebSocket route,
which skipped the API key check.
- The server no longer serves mlx-vlm's `/v1/settings` route, which could
point the prompt cache and the drafter at any folder, or its own
transcription routes, which decoded an upload with ffmpeg before any
check.

### Security

- A request's `chat_template_kwargs` must be an object, and can no longer
replace the model's chat template or set another parameter of the
template call. Such a request gets a 400, so the server never renders
Jinja that a client sends.
- A server, menu bar or extra install that gmlx starts no longer imports a
`gmlx` or `pip` package from the current folder in place of the installed
one. The processes gmlx starts also drop the empty and relative entries of
`PYTHONPATH`, which put the current folder on the import path.
- `gmlx launch` creates a configuration file with mode 600 and keeps the
mode of each file it rewrites, so a file that holds a key stays private.
It writes through a symbolic link that stays inside your home folder
instead of replacing it, and refuses a link that leads outside.
- A server on a loopback address refuses a request whose address it cannot
read, and one that reaches it from another address through a redirect,
such as the one a localhost domain of Apple container adds. It also no
longer takes a client's address from `X-Forwarded-For`.
- `gmlx launch open-webui` runs Open WebUI on `127.0.0.1` only and sets its
`CORS_ALLOW_ORIGIN` to its own address, so other computers cannot reach it
and other web pages cannot read its answers. Pass `-- --host 0.0.0.0` to
listen on every address, and export `CORS_ALLOW_ORIGIN` for another
address, such as that of a reverse proxy.
- `gmlx init`, `gmlx pull`, `gmlx sync-models` and `gmlx rm` keep the mode
of the config file, and `gmlx init` gives a new config mode 600. These
commands and the menu bar's Edit config write into the file that a config
link leads to, instead of replacing the link with a plain file.
- When you press Tab, bash completion no longer runs a command, such as
`$(...)`, that a model id, alias or assistant name in the config holds.
Completion leaves out such a name, so a script from 0.4.19 that you saved
to a file is safe too.
- The menu bar, `gmlx service` and the server run `launchctl`, `ps`,
`sysctl`, `vm_stat`, `codesign`, `open`, `pgrep` and `pbcopy` by their
full paths, never through `PATH`.


## [0.4.19] - 2026-09-27

### Added
Expand Down
2 changes: 1 addition & 1 deletion pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"

[project]
name = "gmlx"
version = "0.4.19"
version = "0.4.20"
description = "A local inference platform for Apple Silicon: run, chat with, serve, and fine-tune the GGUF ecosystem's quantized models natively on MLX, straight off the file."
readme = "README.md"
requires-python = ">=3.11" # 3.10 EOLs 2026-10; mlx-kquant's lower floor is a library floor
Expand Down
Loading