Commands

nolune help prints this list for the version you have. Admins can also ask nolune in a chat to change these (“use Sonnet for new chats”, “add an account for Grandma”): the built-in nolune skill tells it how, and it restarts itself only when asked.

Getting started

nolune setup: interactive first-time setup (your account, and the address the family opens: nolune’s relay, this computer, or your own URL); the model comes after: a new profile’s welcome on the web asks for one, or use key set and preset add below
nolune start: run the gateway in the foreground
nolune service install|uninstall|restart|status|logs [-f]: run it in the background (macOS, or Linux with systemd)

Reaching nolune from anywhere

No tunnel, port forwarding or domain of your own. See Remote access.

nolune relay enable [--name NAME]: get a public address through nolune’s relay, like https://NAME.nolune.family (a random name without –name). It passes the family’s traffic to this computer, and could see it, as any tunnel could. --server URL uses a relay of your own
nolune relay status: the address, whether the gateway is connected, and this month’s traffic
nolune relay disable: stop using the relay and give the address back

Settings

nolune config: show address, port and what’s configured
nolune config set <host|port|origin> <value>: origin = the public URL people open (the relay’s address is used instead while the relay is on)
nolune config set image-model <provider/model>: for pictures, e.g. openai/gpt-image-2.5-flare
nolune config set embeddings <auto|off|provider/model>: what memory search finds meaning with: auto uses the OpenAI key, else OpenRouter’s; the provider is openai, openrouter or custom-openai (a custom provider’s model, like custom-openai/ollama/nomic-embed-text)
nolune config set claude-path <path>: the Claude Code that claude-plan chats run (found on the PATH and in its usual folders otherwise)
nolune config set command-mode <auto|unrestricted>: auto (the default): a model checks each command the agent runs and blocks what could do harm nobody asked for; unrestricted runs them unchecked (not recommended). Not from the agent’s own commands
nolune config set safety-model <preset|chat>: the preset whose model does auto mode’s checks, or chat for each chat’s own model (the default)
nolune config set update-check <on|off>: on (the default): the gateway asks GitHub once a day for nolune’s newest release, and admins see when there’s one
nolune key set <anthropic|openai|openrouter|xai|firecrawl> [key]: store an API key (prompts if omitted) after checking it; OpenAI’s runs GPT chats and makes pictures, xAI’s runs Grok, Firecrawl’s lifts the web search’s daily limit. Admins can also do this on the web, under Models & keys
nolune key rm <anthropic|openai|openrouter|xai|firecrawl>: remove a stored key (the environment’s is used, if set)
nolune env set <NAME> <value>: extra env var for agent commands (e.g. HASS_TOKEN)
nolune env rm <NAME> | nolune env list

Custom providers

Model servers of your own: Ollama, LM Studio, oMLX, vLLM, llama.cpp.

nolune provider add <name> <url> [--api openai|anthropic] [--key K]: add one, through its OpenAI API (the default) or its Anthropic API, or change the address and key of the one of that name; nolune asks it for its models first, and for its key at a terminal when it wants one. The address is the server’s, like http://localhost:11434. A server that speaks both can be added once for each
nolune provider rm <name|id>: presets on it stop working until they’re moved
nolune provider list

Plans

Chats on your own subscription instead of an API key.

claude-plan: a Claude Pro or Max plan, through Claude Code on this computer, signed in to your Claude account; Claude Code keeps the sign-in. chatgpt-plan: a ChatGPT Plus or Pro plan, signed in with ChatGPT (Sign in with ChatGPT), which nolune keeps in ~/.nolune/chatgpt.json. Plan limits assume one person’s ordinary use: keep busy automations and subagents on an API key.

nolune <plan> status: who it’s signed in as (and which Claude Code runs)
nolune claude-plan setup: install Claude Code and sign in, where needed, asking first, in a terminal
nolune chatgpt-plan setup [--another-account]: sign in with ChatGPT in a browser, where needed; from another device, paste the address it ends on. Signs in to the account used last unless told otherwise
nolune chatgpt-plan logout: sign out; chats on chatgpt-plan presets stop until someone signs in again
nolune chatgpt-plan models: the models the plan offers, for nolune preset add

nolune-plan: a subscription to nolune itself, which covers chats, pictures and memory search with no keys at all, on the month’s credits. Linked with a code you approve on nolune’s page, from any device; nolune keeps the link in ~/.nolune/nolune-plan.json. Subscribe at api.nolune.dev.

nolune nolune-plan setup: link nolune to your plan, where needed
nolune nolune-plan logout: unlink; chats on nolune-plan presets stop until it’s linked again
nolune nolune-plan models: the models the plan offers, for nolune preset add

Users

There’s no sign-up page: admins add people here, or on the People page in the web UI.

nolune user create <name> <email> [--password P] [--admin]
nolune user invite [name]: a link to send, where they make their own account (once, within 7 days); name: who it’s for
nolune user passwd <name|email> [--password P]
nolune user admin <name|email> [--off]
nolune user rm <name|email>
nolune user list: name, email, admin, and their picture’s file if they have one

Model presets

Shared by all profiles.

nolune preset add <model> [--provider anthropic|openai|openrouter|xai|claude-plan|chatgpt-plan|nolune-plan|<custom>] [--name N] [--context-window TOKENS]: the provider checks the model id first (anthropic unless given); OpenAI models other than the flagships need --context-window. OpenRouter’s ids name their maker (anthropic/claude-sonnet-5), and the model must be able to call tools. xAI’s are Grok’s (grok-4.7), and it lists them with their windows. A custom provider (by its name) lists its models; its model must call tools too, which shows at its first reply, and pictures and PDFs reach it as paths. The plans check their sign-in instead, and chatgpt-plan the models the plan offers
nolune preset edit <name|id> [--provider P] [--model M] [--name N] [--context-window TOKENS|auto]: change what’s given; a new model is checked like add’s. Chats already on the preset keep what they had
nolune preset rm <name|id>
nolune preset default <name|id>: the model new chats start with
nolune preset list

Profiles and skills

nolune profile list: slug, name and avatar of every profile
nolune profile avatar [<name>] [--profile SLUG]: show or change the assistant’s avatar in the web chat: probe, campfire, lantern, planet, quantum, comet, moon, satellite
nolune skill new <name> [--description D] [--profile SLUG | --global]
nolune skill list [--profile SLUG]
nolune skill enable <name>... [--profile SLUG]
nolune skill disable <name>... [--profile SLUG]: leave out of the profile’s chats

MCP servers

Other apps’ and services’ tools, which chats get next to nolune’s own. See Connected services.

nolune mcp add <name> <url> [--transport http|sse] [--header "Name: value"]... [--client-id ID [--client-secret S]] [--scope S] [--description D] [--profile SLUG]...: connect a server at an address. --client-id: the app a service had you register for its sign-in
nolune mcp add <name> [--env NAME=value]... [--cwd DIR] [--description D] [--profile SLUG]... -- <command> [args...]: connect one this computer runs. nolune connects to check it; the same name again changes it. --profile: only in those profiles (every profile without it)
nolune mcp add-json <name> '<json>': the same, as MCP clients’ settings write a server
nolune mcp rm <name>
nolune mcp list: the servers, the names of their keys (never the keys), and whether someone signed in to them
nolune mcp login <name>: sign in to a server that wants it (OAuth): prints the page to open, which comes back to nolune’s web address
nolune mcp logout <name>: forget its sign-in
nolune mcp tools [<server> [<tool>]]: their tools; with a tool, what it does and takes
nolune mcp call <server> <tool> [<json>|-]: call a tool with a JSON object of arguments (- reads stdin); pictures it returns are attached for the agent

Automations

Results show up as notifications in the web UI.

nolune trigger add <name> WHEN WHAT [--summary S] [--icon I] [--preset NAME] [--effort LEVEL] [--profile SLUG]
- WHEN: --cron "<min hour day month weekday>" (local time) | --at "YYYY-MM-DD HH:MM" | --in 30m|2h|1d | --webhook
- WHAT: --prompt "<what the agent should do>" | --script "<shell command, no model>"
- --summary: one plain sentence the family sees on the Automations page
- --icon: a Lucide icon name for it, like umbrella
nolune trigger list [--profile SLUG]
nolune trigger show|run|pause|resume|rm <name|id>
nolune trigger edit <name|id> [--name N] [--summary S] [--icon I] [WHEN] [WHAT] [--preset NAME] [--effort LEVEL]
nolune wake <message> [--title T] [--profile SLUG]: start a background agent run now; trigger scripts call this (nolune wake - reads stdin)

Memory

Short notes in fixed categories; the agent reads the ones it needs.

nolune memory [list] [--profile SLUG]: the notes, how many facts each holds, and whose they are
nolune memory search <words>...: find facts in every note, best match first
nolune memory show <topic>...: print notes (a topic is home, people/anna, …)
nolune memory add <topic> <fact>: add one fact; the note is created if needed
nolune memory replace <topic> <old> <new>: change text that appears once in the note
nolune memory forget <topic> <text>: remove the one line that contains <text>
nolune memory write <topic> [text]: replace the whole note (the text, or stdin)
nolune memory rm <topic>
nolune memory mv <topic> <new-topic>
nolune memory merge <topic> <into-topic>: put one note into another about the same thing
nolune memory learning [on|off]: whether nolune also saves what it learns by itself, looking over each chat once it goes quiet

Categories: core, people/<name>, home, health, plans, routines, pets, places, projects/<name>, other. Facts go only into these; a note from before them can be read and rewritten until it’s moved. The note core is pinned: every new chat starts with it, so it holds at most 4000 characters.

Each member also has a card, cards/<name>: a note about them that goes with them into all their profiles (at most 2000 characters). show, search, add, replace and forget work on it; a change is refused unless its owner wrote one of the messages the agent is answering.

Cards

nolune card [<name|email>]: print someone’s card, or list everyone’s

People edit their own card on its page in the web app (Your card, in the menu under their name).

Soul

Who nolune is for a profile: character, values, tone; at most 4000 characters.

nolune soul [show] [--profile SLUG]: print it
nolune soul write [text]: replace it (the text, or stdin); every chat gets it from its next message
nolune soul rm: remove it

Pictures

Model openai/gpt-image-2.5-flare; change it with nolune config set image-model.

nolune generate image <PROMPT | -> [--image FILE]... [--size square|portrait|landscape|auto|WxH] [--quality Q] [--background auto|transparent|opaque] [--format png|jpeg|webp] [--count N] [--model PROVIDER/MODEL] [--out DIR|FILE] [--dry-run]: make pictures from a prompt, or change the --image ones; - reads the prompt from stdin

The web

Through Firecrawl: its free tier, limited per day, until nolune key set firecrawl.

nolune web search <query> [--limit N] [--news] [--recent day|week|month|year] [--country CC]: the top results’ titles, addresses and snippets (5 unless –limit, at most 20); –news for news articles, –country as someone there sees them (de)
nolune web read <url> [--out FILE]: a page’s main text (or a PDF’s) as Markdown; a long one is saved whole to a file, whose path it prints

Inside agent commands

NOLUNE_PROFILE is set, so --profile can be left out.

nolune view <image>...: show images to the agent: they’re attached to the command’s result (HEIC and big photos are converted)

Subagents

In agent commands: agents that work on a task in the background, in a conversation of their own that starts with only the task.

nolune agent run [<id>] --prompt "<task>" [--preset NAME] [--effort LEVEL]: start a subagent, or give one that finished more work; prints its id and its log file (--prompt - reads stdin). It uses this chat’s model and reasoning unless given: --preset takes a name from nolune preset list, --effort one of low, medium, high, xhigh, max
nolune agent watch <id>: wait until it’s done and print its last message; run it with run_in_background to be told when it’s done
nolune agent steer <id> --prompt "<message>": message a subagent while it works
nolune agent stop <id>
nolune agent list