Repository navigation
Tags: ModelsLab/modelslab-cli
Tags
v0.2.2 — Homebrew and Scoop get the 0.2.x CLI
Same CLI as v0.2.1. This release exists so the Homebrew cask and the Scoop
manifest publish again: they were stuck on 0.1.4 behind an expired token and
then a release-config bug, both now fixed.
If you installed with Homebrew or Scoop, update now:
brew upgrade modelslab/tap/modelslab
scoop update modelslab
What changed since 0.1.4 for those users is in the v0.2.0 and v0.2.1 notes:
browser login by default, working `modelslab update`, clearer login errors,
and documented exit codes.
FIXED
- Release: the Homebrew and Scoop publish tokens are rendered the way
goreleaser requires, so a valid token no longer fails the release.
v0.2.1 — browser login by default, working updates on Windows and pac…
…kage managers
UPGRADING FROM 0.2.0
- On Windows, `modelslab update` in 0.2.0 cannot install this release. Update
the way you installed:
pip: pip install --upgrade modelslab-cli
npm: npm install -g modelslab-cli@latest
Scoop: scoop update modelslab
Homebrew: brew upgrade modelslab/tap/modelslab
CHANGED
- `modelslab auth login` opens the browser by default in an interactive
terminal. This works for every account, including "Continue with Google" and
GitHub accounts, which have no password. Email and password are still used
with --email, --password or --browser=false, when stdin is not a terminal,
and over SSH or on Linux with no display.
- `modelslab update` no longer replaces a binary that pip, pipx, uv, npm,
Homebrew or Scoop installed. It prints that package manager's command.
FIXED
- `modelslab update` always failed on Windows ("cannot replace a running
executable").
- An empty password printed "password: no input provided" with no hint. It now
points Google and GitHub accounts at the browser login.
- The update notice told every user to run `modelslab update`. It now shows the
command for how the CLI was installed.
- Offline or behind a proxy, every command waited 1.5s on the update check.
- The update notice printed in CI and into captured stderr.
- "permission denied" during an update now says to use sudo or an
Administrator terminal.
v0.2.0 — CLI authentication diagnostics and credential safety BREAKING - Every command now exits with its documented code instead of 1. An auth failure is 3, usage 2, rate limit 4, not found 5, payment 6. Scripts that test for `-eq 1` need updating. - A project-level .modelslab/config.toml no longer supplies base_url, api_key or token. Use --base-url or MODELSLAB_BASE_URL. Other preferences still merge. SECURITY - .modelslab/config.toml is read from the current working directory and was merged wholesale, so a committed base_url in any repository you cloned and cd'd into redirected the next command — with your stored bearer token in the Authorization header — to that host. FIXED - Failed logins printed the raw JSON error envelope instead of the message. - The login hint said "check your email and password" to Google and GitHub accounts, which have no password at all; it now points at --browser. - A 502 during a deploy was reported as invalid credentials. - `auth login` printed "Logged in" even when the credential write failed. - The authorize URL was never shown when the browser failed to open. - Piped and CI logins died on a TTY check, leaving --password on the command line as the only option. - The MCP server's auth-login returned a token without applying it, so every later tool call in the process stayed unauthenticated. - 429 now honours Retry-After and keeps the server's own message. - Prompts read a whole line and reject empty input.
v0.1.4 Completes npm and PyPI distribution. npm install -g modelslab-cli pip install modelslab-cli npm platform binaries ship as @modelslab/cli-<os>-<arch> optional dependencies; the entry package stays unscoped. Wheels are per-platform with a console script.
v0.1.3 Completes the npm and PyPI distribution that v0.1.2 published only partially. npm platform packages are scoped as @modelslab/cli-<os>-<arch>; the entry package stays modelslab-cli. PyPI publishes independently of npm, and both publishers skip what is already on their registry so a partial release can be recovered by re-running. npm install -g modelslab-cli pip install modelslab-cli