Use litecode

Commands

Shell commands#

Command	What it does
`litecode`	interactive session in the current folder
`litecode "TASK"`	run one task, then exit
`litecode setup`	wizard: folders, engine, models
`litecode tune [ROLE...]`	measure speed, save best threads and speculative mode
`litecode eval [NAME...] [--role R]`	run 8 real agent challenges and score them
`litecode doctor`	hardware, config, engine, installed models
`litecode init`	create `LITECODE.md`
`litecode scan [PATH] [--json]`	static security scan, no model
`litecode review [PATH]`	scan, then model verdict per finding
`litecode plan "TASK"`	read-only plan
`litecode skills`	list skills
`litecode skills new NAME`	create a skill step by step
`litecode skills rm NAME`	delete a user or project skill
`litecode mcp`	list MCP servers (`--tools` to list their tools)
`litecode mcp add NAME CMD [ARGS...]`	add an MCP server to the user config
`litecode mcp rm NAME`	remove it
`litecode learn export FILE [--all]`	write a fine-tune dataset (rated tasks; `--all` adds unrated)
`litecode config`	show all settings
`litecode config KEY VALUE`	set one setting (`roles.code NAME.gguf` for roles)
`litecode fim`	autocomplete server for editors
`litecode --version`	version

Flags: --yes (approve all), --readonly, --role NAME, --plan.

Command	What it does
`/help`	list commands
`/plan TASK`	plan with no edits, then ask to run it
`/review [PATH]`	security review
`/scan [PATH]`	static scan only
`/test [CMD]`	run tests; if they fail, fix and run again (max 3 rounds)
`/diff`	`git diff`
`/commit`	model writes a commit message; you confirm before commit
`/model [ROLE]`	show roles, or force one (`auto` = router decides)
`/mode [ask\|auto\|readonly]`	permission mode
`/remember TEXT`	save a project fact
`/prefer TEXT`	save a preference for all projects
`/memory`	show preferences and facts
`/forget [facts\|prefs]`	delete memory (both if no argument)
`/good`, `/bad`	rate the last answer
`/undo PATH`	restore the newest checkpoint of a file
`/sessions`	list sessions in this project
`/resume [ID]`	continue a session (latest if no ID)
`/clear`	start a new session
`/init`	create `LITECODE.md`
`/skills`	list skills
`/skill new NAME`	create a skill step by step
`/skill make DESCRIPTION`	model drafts a skill; you confirm
`/skill rm NAME`	delete a skill
`/NAME [TASK]`	run skill NAME
`/mcp`	list MCP tools
`/config [KEY VALUE]`	show or set config
`/cost`	tokens and model time this session
`/exit`	quit (saves cache, extracts facts)

Key	Default	Meaning
`engine`	`<home>/engine/llama-server`	llama.cpp server binary
`models_dir`	`<home>/models`	GGUF folder
`data_dir`	`~/.litecode`	cache, slots, scratch, memory, skills
`project_memory`	`project`	`project` or `central`
`threads`	4	CPU threads
`ctx`	8192	context tokens
`kv_type`	`q8_0`	KV cache type
`cache_ram_mb`	1024	server prompt cache RAM
`gpu_layers`	0	layers on GPU (GPU builds only)
`speculative`	`auto`	`auto`, `off`, `ngram-simple`, `ngram-mod`
`kv_persist`	true	save prompt cache per project
`permission`	`ask`	default permission mode
`auto_learn`	true	extract facts at session end
`lora`	empty	LoRA adapter path
`max_steps`	24	tool steps per task
`temperature`	0.1	sampling temperature
`port`, `fim_port`	8080, 8081	local ports
`roles.<role>`	see config	model file per role: code, chat, fim, security, quality