The ModelBound CLI — token optimization, skill pipeline, and version management from your terminal.

modelbound lets AI engineers run the same optimization and Skill Development Pipeline that powers the ModelBound web app — directly from their shell, CI, or git hooks. No web UI required.

npm i -g modelbound
modelbound auth login
modelbound optimize ./skills/code-review/SKILL.md --dry-run
modelbound pipeline run code-review --apply-optimization --bump patch

ModelBound's job is to make AI agents more reliable and cheaper to run. Most of that work happens inside an editor, an MCP server, or a CI step — not a browser. This CLI is the lowest-friction way to:

• Cut tokens on a skill, rule, or whole repo with one command
• Run the full Skill Development Pipeline (Test → Optimize → Production) from CI
• List, diff, and restore versions without leaving the terminal
• Sync skills between your repo and your ModelBound team

Every command that mutates a local file writes a backup first:

.modelbound/backups/<relative-path>/<iso>-<sha7>.bak

Every summary line includes the one-command restore path. Cloud writes use the existing file_versions table — the version id is printed alongside the summary so you can roll back with modelbound version restore.

pipeline run --apply-optimization defaults to a dry-run + confirmation unless --yes is passed. Gates (trust, latency, tests) are never bypassed without explicit --override-gates.

See docs/commands.md for the full reference. Quick tour:

modelbound auth login                # device-code OAuth
modelbound detect                    # detect IDE layouts (cursor, claude, copilot, …)
modelbound optimize <path|skill-id>  # streams progress, writes backup
modelbound pipeline run <skill-id>   # full Test → Optimize → Production pipeline
modelbound version list <skill-id>
modelbound version restore <skill-id> <version>
modelbound push <path>               # local → cloud
modelbound pull <skill-id>           # cloud → local (backup first)
modelbound mcp print-config          # spit mcp.json for current IDE

Global flags: --json, --quiet, --no-color, --profile <name>.

modelbound config stores per-profile settings in ~/.config/modelbound/config.json (via conf):

modelbound config set apiUrl https://api.modelbound.co
modelbound config set defaultIntensity balanced
modelbound config set autosync true

Auth tokens are stored in the OS keychain when available, falling back to the same conf file (0600).

This CLI shares its core (API client, backup engine, progress renderer) with modelbound-mcp. Anything you can do here you can also do via MCP tools (optimization.*, pipeline.*, skill.*). Pick whichever entrypoint fits your workflow — the safety guarantees are identical.

Install hub: modelbound.co/connect

PRs welcome. Please open an issue first for anything non-trivial. CI runs lint, typecheck, and unit tests on every PR; security scans (Dependabot, CodeQL) run weekly.

MIT