Lag: Lamin's Agent CLI

lag can retrieve or generate scripts (.py), notebooks (.ipynb), and skills & plans (.md) and then perform work with these tools.

Setup

pip install lag-cli

Create ~/llms.env with:

GEMINI_API_KEY=your_api_key_here

Make sure LaminDB is initialized and connected.

Quick Start

Run in default mode:

lag --prompt "Write a text file with 'Hello agent!' in it, please"

Modes

Default mode

• If a plan exists (plan.md or latest plan_*.md, or --plan-file), lag executes the referenced runnable tools.
• Otherwise, lag generates runnable tool file(s), shows their contents, and asks for confirmation before execution.
• In default mode, generated code includes ln.track() / ln.finish() unless --no-track is used.

Planning mode (--plan)

Generate tools without executing them:

lag --plan --prompt "Write a plan for this analysis"

Generated tool files are saved via lamin save in plan mode.

Common Flags

• --project <name> sets LAMIN_CURRENT_PROJECT.
• --model <model-name> selects the Gemini model.
• --output-file <path> sets output filename for generated content.
• --output-format [md|py|ipynb] applies in --plan mode.
• --plan-file <path> executes a specific plan in default mode.
• --no-track disables ln.track() / ln.finish() injection.
• --yes skips confirmation for executing newly generated tools.

Run Context Propagation

When lag executes scripts/notebooks, it propagates:

• LAMIN_INITIATED_BY_RUN_UID (master run uid)
• LAMIN_CURRENT_PROJECT (if --project is provided)

lag itself does not directly create output artifacts; produced outputs are tracked by executed tool code.