Glossary

Plain definitions of the words you'll meet in Murage, in alphabetical order.

A#

ACP engine: an engine Murage runs through the Agent Client Protocol, a standard for running other agents. Some app messages mention it, such as "Choose Claude or an ACP engine."

Allow for this task: an approval that covers the same kind of action in the same place until the current task ends. Offered on stop-line cards.

Allow once: approve one action, this time only.

Always allow: stop asking about this. On a stop-line card it's tied to that folder, recipient or payee. On an ordinary connected-app card it covers the whole tool.

Approval card: the message a bot shows when it needs your permission, with buttons to allow or deny.

Approval level: how often a bot stops to ask you. One of Ask, Auto, Full access or No limits.

Archive (memory): stop using a memory while keeping its sources for history.

Ask: the approval level where a bot asks before any action that needs permission. New bots start here.

Auto: the approval level where a bot keeps going on its own, and destructive or sensitive actions still ask.

B#

Blocked: a Skill Guard verdict. The skill can't be switched on.

Bot: one member of your Murage team, with its own name, instructions, engine, memory, skills and working folder.

Box: a paid cloud computer service from ascii.dev that can give a bot its own Linux desktop. Used with your own Box key.

Built-in browser: a web browser inside Murage that a bot drives while you watch. An experimental feature, on Mac and Ubuntu.

C#

Capture and recall: the default memory mode. Murage records conversations and brings relevant memories into later turns.

Channel (room): a group chat where you and several bots work together, often with a lead.

Channels (messaging apps): Telegram, Slack and Discord, when paired so you can message your Chief of Staff from them.

Checkpoint: a snapshot of a bot's files taken before each turn.

Chief of Staff: the one bot you talk to that builds your team, hands out work and reports back. Named Ember by default.

Connected apps: outside services such as Gmail, Slack, Notion or GitHub that bots can use as tools. They come through Flux Router's connected apps or your own connected-apps key. Up to five labeled accounts per app.

D#

Deny: say no to an approval card. The bot is told and can try another way.

Diagnostics: a text file of versions, on/off settings and a redacted log tail, made with Export Diagnostics… for bug reports.

E

EmberBot Markdown: the readable Markdown version of a team package, with a .emberbot.md name.

Engine: the AI program that does a bot's thinking and acting, such as Claude Code, Codex, Fuigo, Hermes or a local model. Each bot has one.

Episodic memory: memory of what happened. Conversation excerpts and their sources.

F

Flux Router: an optional paid service that gives one key for 70+ models, connected apps, voice, web search and images.

Forget (memory): remove a memory or source for good, and anything built from it.

Fuigo: Murage's built-in agent engine, installed with the app. It needs a model, such as a Flux Router or provider key.

Full access: the approval level where a bot doesn't ask, except at the stop line and before reading your keys and passwords.

H

House Rules: your rules every bot reads first, in every conversation, like "never quote a price". Edited in Settings → House rules.

I

Identity memory: memory of who's who, including you and the bot itself.

Inbox: the one place for approvals, questions, decisions, failures and reports that need you. Also shown as Needs you in the sidebar.

L

Local model: an AI model running on your own computer through Ollama, LM Studio, llama.cpp or similar.

Local VM: a Linux desktop in a container on your own computer, given to one bot.

M

MCP server: a plug-in that gives bots new tools, using the Model Context Protocol standard. You can add your own, and other AI apps can use Murage as one.

MEMORY.md: each bot's own notebook file, read on every turn.

Menu bar / system tray: Murage's icon with a count of what needs you, and quick approvals.

N

Needs a look: a Skill Guard verdict. The skill only switches on after you choose Use it anyway.

No limits: the approval level where a bot does anything without asking except reading your keys and passwords. The stop line is off. Desktop only.

No red flags: the clean Skill Guard verdict. Nothing risky matched; not a guarantee.

P

Package: a team definition in the murage.package format, shared as JSON, ZIP or EmberBot Markdown.

Pairing: linking your phone, Telegram, Slack or Discord to Murage with a one-time code.

Personality: a short description of how a bot talks, sent to the model with every turn.

Pin (memory): keep an important memory at the front of recall.

Playbook: instructions in a team package that load only when a message matches their trigger words.

Procedural memory: memory of how to do things. Can be turned into a skill.

Project: a channel with a goal, its own files and a status.

R

Routine: work that runs on a schedule, once, daily or every so many minutes. Runs only while your computer is on and Murage is running.

S

Scope: the place a memory belongs to: a conversation, bot, channel, team, project, the workspace or your preferences.

Semantic memory: plain facts.

Skill: a written procedure a bot can follow, stored as a SKILL.md file.

Skill Guard: Murage's scanner that checks every skill for risky patterns.

Stop line: the few actions that wait for your yes, unless you have allowed them for that task or switched the stop line off with No limits: messaging anyone new, posting publicly, paying, or deleting outside its folder. It applies on Ask, Auto and Full access. On Hermes and other ACP engines, it covers only what the engine asks Murage about.

T

Team: a group of bots with a lead, shown together in the sidebar.

Team brief: instructions you write for a team that every member reads.

Team library: Murage's collection of ready-made team packages.

Team map: the view of your Chief, team leads and specialists, and where you manage memory.

This computer: the Computer panel choice that lets a bot use your own screen. Mac, and Ubuntu on Xorg.

Turn: one round of a bot working on a message, from your message to its finished reply.

U

Use my Chrome: letting one bot use your own signed-in Google Chrome.

W

Watch: a routine that re-checks files in a bot's working folder and reports what changed.

Webhook: a local address another app can call to start a task for a bot.

What shapes a bot: the settings view that lists everything a bot reads, in order, with Show exactly what it read.

Working folder: each bot's own folder for its files. The stop line guards deletes outside it.

Working memory: memory of where work stands right now.

/learn: the command that has a bot write a new skill from a description, link, folder or what you just did. Experimental; the skill lands switched off.