CLI documentation

LOLM is a local command-line agent. The website is documentation only.

Quick start

npm install -g lolm-cli
lolm setup
lolm doctor
lolm

Commands

Command	Purpose
`lolm`	Persistent customer console with conversation and slash commands.
`lolm ask`	Answer with read-only file and web tools.
`lolm code`	Inspect, edit, execute, and verify in a working directory.
`lolm pdf`	Generate, validate, and write a real local PDF.
`lolm html`	Write one self-contained HTML document.
`lolm try again`	Resume the saved prompt, output path, and working directory.
`lolm update`	Check npm and update the installed CLI.
`lolm doctor`	Verify runtime, provider, key, model discovery, and NFET assets.

Natural text opens the persistent console in a real terminal. Add --once for scripts or one response.

Providers

lolm providers
lolm setup anthropic
lolm models
lolm ask "hello" --provider openai --model MODEL

Adapters support OpenAI-compatible APIs, Anthropic Messages, Gemini Generate Content, and Ollama Chat. Presets cover OpenAI, Anthropic, Google, xAI, OpenRouter, Groq, Mistral, DeepSeek, Together, Cerebras, Ollama, and custom endpoints.

NFET control

The persistent Python bridge loads the frozen local backbone and trained graft once. Candidate output is replayed through the exact model formulas. Rolling calibration and the trained head produce continue, retrieve, verify, branch, or finalize. The agent follows those decisions and displays their source.

export LOLM_HOME=/path/to/LOLM
lolm nfet status
lolm nfet test "text to monitor"
lolm config set nfet-profile qwen3_4b_lab
lolm config set nfet-device mps
lolm config set nfet-checkpoint runs/nfet_controller/live_qwen4b.pt

Use --no-nfet only when you intentionally want provider-only operation.

Local safety

Reads are bounded and exclude common generated directories.
Writes and shell commands ask first unless --yes is supplied.
Broad destructive commands remain blocked.
--dry-run previews changes.
Remote URL fetches reject private and loopback targets.
API keys are stored locally and never printed.
Remote model providers receive the prompt and selected context used for inference; choose Ollama for local inference.

JSON mode

lolm doctor --json
lolm code "run tests" --json --yes

Stdout contains exactly one JSON document. Errors use {"ok":false,"error":{"code":"…","message":"…"}}.

Configuration

lolm config show
lolm config set provider openrouter
lolm config set model MODEL
lolm config unset api-key

Configuration is stored at ~/.lolm/config.json with mode 0600. Environment variables take precedence. See the source repository for architecture details and commercial licensing.