Settings → Commands

A command is a short phrase that the app runs itself. You can say it to the orb or type it in the chat. The same phrases work both ways. The rail entry is Commands. The page subtitle reads The phrases the app runs itself, said to the orb or typed in the chat.

The page has two tabs: Phrases and Routines. The app remembers the last tab you opened.

For the voice side (the orb, the wake word, the microphone), see Settings → Voice and Calliope, the voice.

Phrases tab

Already understood

This card lists the commands the app knows out of the box. The list of windows comes from the live catalogue, so a new widget or an installed app shows up here without an update of the page.

• “open”, “show”, “launch” + a window name, followed by the names of every window you can open.
• “close …”, “fullscreen”, “normal size”: on the named window, otherwise on the last one used.
• “new chat”, “grid”, “governance”.
• “add a task …” writes to the backlog. Typed, a request for an appointment or a task opens a proposal to approve.
• A command is 7 words at most. Politeness and punctuation are ignored. A command can open a longer sentence: “open the agenda and tell me …”.

Under the list, a hint says that the model knows this list and your taught phrases. You can ask it “what are my commands”.

Custom commands

You teach a phrase and pick one action for it. A taught phrase is read before the built-in commands. The hint reads A phrase triggers an app action, spoken or typed. It comes before the other commands. Punctuation and polite words are ignored.

1. Type the phrase

   Type it in the field The phrase to say or type.

2. Pick the action

   Open Choose the action… and pick one entry. The list has four groups, described below.

3. Teach it

   Press Teach or Enter. The phrase appears in the list above the field, as « phrase » → action.

The four groups of actions:

• Open a window: every window of the catalogue, built-in widgets and installed apps.
• The current window: Close the current window, Put the current window in fullscreen, Restore the current window.
• System: the actions of the keyboard shortcuts list, such as a new chat, the grid or governance. The command palette and "close window" are left out of this group. See Settings → Shortcuts.
• Programs on this computer: the programs of your system's launcher. While the list loads, the group shows Reading programs…. If the system refuses the read, it shows Programs could not be listed on this machine. An empty list shows No program found.

Every action comes from a list the app owns. You cannot type a free command or a file path.

Rules the app applies:

• A phrase taught twice keeps only the last action. One phrase, one action.
• A phrase already used by a routine is refused. The message reads Phrase already used by the routine “name”.
• The cross button next to a row (tooltip Forget this phrase) removes it.
• With nothing taught, the card says No phrase taught yet.
• A failed write shows The phrase could not be saved: it will be forgotten at the next start.

Custom commands are stored in the app's local storage.

The assistant

A sentence that starts with the assistant's name hands it a task. The name is the one you gave the assistant during onboarding or in Settings → Profile. The default name is Mnemosyne. In the card text below, name stands for that name. The card reads “name, find the museum's opening hours”. The assistant shows what it will do and waits for your click.

On that approval card, a checkbox such as Stop asking for web searches stops the questions for one kind of task. Three kinds can be trusted this way: Web searches, Summaries and Images. An action that leaves your machine is asked every time.

This section lists the kinds you trusted:

• With none trusted, it reads name asks before every task.
• Otherwise it reads name starts without asking: followed by each kind.
• Ask again next to a kind puts the question back for that kind.
• A failed write shows This choice was not saved: the question will come back.

A trusted task can cost money

The approval card warns that one task can cost up to about $0.50. The daily token cap stays on.

Routines tab

A routine is a name, a phrase and a list of steps run in order. You say or type the phrase, and the app runs every step. Each routine is one card in a grid.

The first time you open the Commands page, the app adds two example routines, both Off:

• Work mode (phrase work mode): opens Agenda, To-do and Notes.
• Evening mode (phrase evening mode): closes every window of the desktop, then opens Agenda.

The examples are offered once. If you delete them, they do not come back. The step labels of the examples (Agenda, To-do, Notes) are hardcoded in English in the code, whatever the app language.

Create a routine

1. Add a card

   Press New routine. A card appears, named New routine, switched on, with no phrase and no step.

2. Name it and give it a phrase

   Edit the name field at the top of the card. Type the phrase in the Phrase field (placeholder the phrase to say). Both save when you press Enter or leave the field.

3. Add the steps

   In the step row, pick a kind in the first list, then a target in Choose…, then press Add.

4. Try it

   Press Try to run the routine now and see what each step did.

You can keep up to 30 routines. Each routine holds up to 12 steps. The New routine button greys out at 30, and Add greys out at 12 steps.

The card

• The name field: the routine's name. An empty name goes back to the previous one.
• On / Off: the switch. Only a routine switched On answers to its phrase.
• Phrase: the phrase that runs it. Matching is exact on the whole phrase, with politeness and punctuation folded out, like custom commands. A phrase already used elsewhere is refused under the field:
  • Phrase already used by the command “label”.
  • Phrase already used by the routine “name”.
• With no phrase, the card says No phrase: runs from here only. The Try button still runs it.
• The steps, numbered. Each step has Move up, Move down and Remove the step buttons. With no step, the card says No step.
• Try runs the routine now, even when it is Off. It greys out while the routine runs and when there is no step. If the routine needs a confirmation, the card says name launches …: to confirm on screen. The confirmation dialog then opens.
• Delete asks twice. The first press turns it into Confirm for 4 seconds. A second press in that time deletes the routine.
• A failed write shows Not saved: storage refused it.

Step kinds

Each step names something the app owns. The only text you type is a playlist name.

Kind	Target	Step label
Desktop	one of your desktops	Go to the … desktop
Open	a window of the catalogue	Open …
Board	a board an installed app declares	Open the … board
Program	a program of your system's launcher	Launch …
Playlist	a playlist name, typed (120 characters at most)	Play the playlist “…”
Close	a window, or Every window of this desktop	Close … or Close every window

The Playlist kind appears only when MnemoPlayer is installed. MnemoPlayer looks up the name among your own playlists when the routine runs.

While a list of targets loads, the row shows Reading…. If the list cannot be read, it shows List unreadable. An empty list shows No desktop, No program or No declared board.

What a run reports

Try prints one line for the whole run. Then it prints one line per step, marked with a check mark or a cross.

• name: every step ran.
• name: stopped at step 2/4 (step), reason. The routine stops at the first step that fails.
• A step that had nothing to do is marked already done.

A failed step gives its reason in plain words, for example desktop not found, not installed, window not found, no playlist by that name or program refused. The full list of reasons lives in the app's locale files under routines.refused.

Confirmation before a program launches

A routine that launches a program asks first. The dialog title reads Launch name? and the text names the programs: It launches … Asked once per program. Adding a new program to the routine asks again for that program.

• Launch confirms and runs the routine. The dialog then shows the report.
• Cancel, the cross button (Close) and the Escape key run nothing.
• A click outside the dialog does not close it.
• If the routine changed while the dialog was open, nothing runs. The dialog shows The routine changed: here is the list as it is now. It asks again.

Routine files from an agent or an IDE

Below the routines, the card Proposed by a file lists routine files that an agent or an IDE wrote. The hint reads An agent or an IDE writes a file in … Once adopted, it needs a phrase and its switch on. The folder is ~/.mnemosyne/routines. The hint shows its real path on your machine.

The app only reads this folder. It never writes into it.

• Each readable file shows its title, its file name, its summary and its steps.
• Adopt copies the file into a new routine. The new routine is Off and has no phrase. You give it a phrase and switch it on.
• A file that cannot be read is listed with its error, so the agent or the IDE can fix it.
• A step of this file cannot be used: one step is invalid. Adopt is greyed out.
• Read the folder again reloads the list.
• With nothing to adopt, the card says No file to adopt.
• The app reads .json files only, 50 at most, sorted by name. Past 50, the card says how many were left unread. A file over 64 KB is refused unread.

A file can also call actions that an installed app declares. These steps only exist in adopted files. Before such a routine runs for the first time, the confirmation dialog lists the actions: It acts in the apps: … Asked again for each new version of its file.

After adoption

An adopted routine runs its own copy. Editing the file changes nothing until you take the new version. The card shows From the file … and then one of these lines:

• The file changed since the adoption. The new steps are shown below it. Take the new version replaces the steps. The name, the phrase and the switch stay yours.
• The file cannot be read anymore: … The routine keeps its copy.
• The file was deleted. The routine keeps its copy.

A routine file uses the mnemo-steps/1 format, the same format the mnemo steps command of the CLI runs. A routine file also needs a title, which becomes the routine's name.

If a step of the new version cannot be used, the card says This version cannot be taken: a step cannot be used. Fix the file, then read the folder again.

Routines are stored in the app's local storage, like custom commands.