CLI reference

This page is verified against the clap-style argument parser in crates/lime-cli/src/cli/parser.rs. Anything not listed here is not accepted.

Synopsis

lime [OPTIONS] [SUBCOMMAND] [ARGS...]

Run lime with no arguments to start the interactive REPL.

Top-level flags

These flags work without a subcommand and shape REPL behavior.

Flag	Type	Default	Description
--help, -h	bool	—	Print help and exit.
--version, -V	bool	—	Print version and exit.
--model SLUG	string	gpt-5.4	Active model. Accepts any slug from the merged catalog (built-ins + ~/.lime/models.json).
--permission-mode MODE	enum	default	One of default, plan, acceptEdits, bypassPermissions, dontAsk.
--dangerously-skip-permissions	bool	false	Force bypassPermissions.
--settings FILE_OR_JSON	path | inline JSON	—	Extra settings — file path or inline JSON, applied last (highest precedence).
--setting-sources LIST	csv	all	Restrict the merged config to a CSV of policy, user, project, local, flag.
--add-dir DIRECTORY	path (repeatable)	—	Additional directories the agent may read/write (in addition to the workspace).
--continue, -c	bool	false	Resume the most recent session.
--resume [REF]	path | id | latest	latest	Resume a specific session. Trailing tokens after --resume must be slash commands.
--allowedTools TOOLS	csv (repeatable)	all	Restrict the enabled tool set.
--reasoning-effort LEVEL	enum	medium	One of none, low, medium, high, xhigh.
--verbosity LEVEL	enum	medium	One of low, medium, high.
--service-tier TIER	string	—	Provider service tier (e.g. default, priority on OpenAI).
--output-schema FILE	path	—	Constrain the final response to a JSON Schema. See Structured output.
--ide MODE	auto | off | port	auto	IDE companion mode. auto discovers a running VS Code companion via ~/.lime/ide/.

Flags that do not exist

The CLI parser does not accept --output-format, --print, or -p. There is also no prompt subcommand. See Non-interactive use for the supported non-interactive entrypoints.

Subcommands

lime init

Bootstrap repo-local Lime files: AGENTS.md, .lime/, .lime.json, and relevant .gitignore entries.

lime status

Print a snapshot of session, model, token, workspace, and provider state.

lime sandbox

Print the sandbox isolation snapshot. See Sandboxing.

lime help, lime version

Aliases for --help and --version.

lime bootstrap-plan

Print the bootstrap-phase skeleton used by lime init.

lime agents [args...]

List or manage configured agents. With no args, lists all.

lime skills [args...]

List or manage discovered skills.

lime command|cmd|commands [args...]

Manage user-scope custom slash commands at ~/.lime/commands/*.md. Subcommands include list, show, create, edit, delete, help.

lime system-prompt

Render the resolved system prompt as it would be assembled for a turn.

Flag	Default	Description
--cwd PATH	current dir	Pretend the cwd is PATH.
--date YYYY-MM-DD	today	Pretend “today” is the given date.

lime login

Store provider credentials under ~/.lime/. Keys stored here override the corresponding environment variables.

Flag	Description
--status	Print credential status. Combine with --provider to filter.
--provider NAME	One of openai, anthropic, gemini, openai-compat (alias openai_compat).
--service NAME	OpenAI-compatible service name. Namespaces credentials under openai_compat:<service>.
--with-api-key [VALUE]	Supply an API key. If VALUE is omitted, Lime reads the key from stdin (no echo).
--with-base-url URL	Custom endpoint URL. Requires --with-api-key.

lime login --with-api-key                               # OpenAI
lime login --provider anthropic --with-api-key
lime login --provider openai-compat --service groq \
           --with-api-key --with-base-url https://api.groq.com/openai/v1
lime login --status

lime logout

Remove stored credentials.

Flag	Description
--provider NAME	Clear one provider’s credentials.
--service NAME	Clear one compat service.

With no flags, clears all.

lime model | lime models

Manage user entries in ~/.lime/models.json. See Custom models for the full discussion.

Subcommand	Description
list, ls	Print the merged catalog (built-in + user).
add (with flags below)	Add or override a model entry.
rm, remove, delete SLUG	Remove a user entry. Built-ins always survive.

lime model add flags:

Flag	Required	Notes
--slug SLUG	✓	Identifier used by --model / /model.
--provider NAME	✓	One of openai, anthropic, gemini, openai-compat, custom.
--context-window N	✓	Full context window in tokens.
--service NAME		Required when --provider openai-compat. Pins to a credential bucket.
--display-name STR		Defaults to the slug.
--description STR, --desc STR		Defaults to "Custom model".
--api-id STR, --api-model-id STR		Wire model id when it differs from the slug.
--max-output N, --max-output-tokens N		Default response cap. Derived if omitted.
--upper-max-output N		Upper retry / thinking cap.
--auto-compact N, --auto-compact-threshold N		Token count at which auto-compaction fires.
--param-preset NAME		One of none, verbosity, reasoning, reasoning-and-verbosity.
--vision		Model accepts images.
--reasoning-summaries		OpenAI-style reasoning summaries.
--parallel-tools		Multiple tool calls per turn.
--extended-thinking, --thinking		Anthropic-style extended thinking.

lime mcp

Manage MCP servers. Scopes: local (project, untracked), project (committed .lime.json / .mcp.json), user (global ~/.lime/).

Subcommand	Notes
serve	Run Lime itself as an MCP server.
list, ls	List configured servers across scopes.
get NAME	Print one server’s resolved config.
add	Add a server. Flags: --transport stdio|sse|http (-t), --scope local|project|user (-s), -e KEY=VALUE (stdio env), -H "Key: Value" (remote header). End flag list with --. The CLI shortcut covers the three common transports; for ws, sdk, streamable-http, or lime-proxy use add-json.
add-json NAME JSON	Add a server from a literal JSON object — accepts every mcpServers.type value. Flag: --scope. Example: lime mcp add-json my-srv '{"type":"http","url":"…"}' -s user.
remove, rm	Remove a server. Flag: --scope.
enable NAME	Enable a previously-disabled server.
disable NAME	Disable a server without removing its config.
reset-project-choices	Forget per-project trust decisions.

lime bridge

Run the JSON-RPC 2.0 bridge.

Flag	Description
--ws	Switch to WebSocket transport (default is stdio).
--ws-port PORT	Port for the WS bind (default 8765). Implies --ws.
--ws-token T	Shared bearer token for WS auth.
--device-auth	Use the device-registry for WS auth. Mutually exclusive with --ws-token.
--no-redact	Disable outbound redaction for trusted local-only setups.
--status	Print bridge engine status and exit.

lime remote

Drive Lime over a Cloudflare tunnel. See Remote control.

Subcommand	Flags
start	--persistent NAME, --url URL (require each other), --no-auto-install.
pair	--label LABEL (required). Mints a one-time device token.
devices, list	List registered devices.
revoke DEVICE_ID	Revoke a device.

Direct slash invocation

Slash commands accepted on the CLI must come from a lime --resume form. Bare lime /<command> works for a small handful that are non-interactive (/help, /agents, /command, /skills); the rest will print a clear error pointing you at lime --resume.

Exit codes

Code	Meaning
0	Success.
1	Argument parse error or unrecoverable runtime error.
2	Provider authentication failure.
3	Permission denied (in dontAsk mode with a deny rule).
130	Interrupted (Ctrl-C).