Performance: slow bots and long chats

• Pick a faster model for simple work. Big reasoning models are slow on purpose. In Bot settingsModel, give quick jobs a faster model or a lower effort setting.
• Check the engine on its own. If the same engine is slow in its own terminal, the delay is upstream: the provider is busy or your plan is rate-limited.
• Check your plan's limits. Subscription engines slow down or pause when you reach a usage limit. The error message usually says so.
• Local models depend on your hardware. A model that barely fits in memory is very slow. Try a smaller one, and make sure it's loaded with enough context (Murage warns before a turn when the prompt would take most of it).

Things every bot reads on every turn add up:

• House Rules: keep them to what really matters.
• Skills: each enabled skill adds a line to the bot's skill index. Keep each bot's skills to its job.
• Pinned memories: pin only what must always be in front.
• Connected apps and MCP servers: each one adds tools the model has to read. Give bots only the apps they need.

Bot settingsWhat shapes this bot the bot shows everything it reads, and Show exactly what it read shows the last turn in full.

• Murage loads long conversations 100 messages at a time as you scroll, so the app itself stays quick.
• The engine, though, may be carrying the whole history. Very long conversations cost more tokens and can make answers slower and vaguer.
• Start a new conversation for a new job. Memory carries the important facts across.
• On engines that offer it, /compact in the composer's / menu condenses the conversation. The menu shows only commands the bot's engine supports.

• If the same step keeps failing, the default House Rules tell the bot to stop after two identical failures and say what's blocking it. Murage also notices when a bot repeats the same tool call.
• Check the Inbox. The bot may be waiting for your approval or an answer.
• Choose Cancel turn on its card, or stop the turn from the chat, and ask again with clearer instructions.

Every member who answers takes a turn. To speed up a channel:

• Mention the bot you want (with @) instead of asking the whole channel.
• Keep channels to the bots the job needs.
• Channel turns have a maximum duration, set in SettingsBotsAdvancedChannel turn timeout. The default is 5 minutes. Raise it for long jobs, lower it to stop runaway turns.

A routine won't stack copies of itself: a scheduled run is skipped while the last one is still going. If routines are slow, give them smaller jobs or space them out.

• Close conversations and panels you're not using, especially live browser or computer views.
• Local VM desktops use real memory and CPU. Stop the ones you're not using.
• The local memory model, indexing and several Local VMs at once can compete for memory on smaller computers.
• Restart Murage if it has been running for a long time with many busy bots.