Bot settings
What a bot's settings actually change, and what they do not.
Model
| Setting | Effect |
|---|---|
| Provider | OpenAI or Anthropic. Determines which API key is used. |
| Model | Which model answers. This is the setting with the largest effect on cost and quality. |
| Temperature | How varied the replies are. Lower values (toward 0) give more consistent, repeatable answers; higher values give more varied ones. Applied to every reply; unset means 0.7. |
Model choice still dominates behaviour — reach for temperature to adjust a model you are otherwise happy with, not to rescue one that is wrong for the job.
Settings that are refused rather than ignored
A deliberate design choice runs through the rest of the configuration: where there is no mechanism behind a setting, the system rejects it instead of accepting and dropping it. Maximum response tokens, persistent memory, per-agent budgets, behaviour blocks, structured output schemas and tool constraints all fall into this group — see AI agents for the full list.
You will not usually meet these through the bot form. They matter when someone edits an agent definition directly and gets a hard error; the error is the system working as intended.
Context and archival
These do work, and they are the settings most worth tuning:
| Setting | What it changes |
|---|---|
| Window size | How many recent messages the bot is given. Default 50. |
| Max window | The ceiling a more specific setting may raise it to. |
| Hot retention (days / messages) | When messages move to cold storage. Defaults: 30 days, 10,000 messages. |
| Archival enabled | Whether this bot's conversations are archived at all. |
| Include events in context | Whether channel events, not just messages, form part of the history. |
Anything left unset falls through to the tenant setting, then the global one. The page marks which values are inherited rather than set on this bot, which is the quickest way to see what you have actually overridden.
Full behaviour is in Context & memory.
Capabilities and tools
What a bot can do comes from the agents it can route to and the tools those agents may call, not from a checklist on the bot. To change it, change the agent definitions — see AI agents.
Verbosity
How much a bot narrates its intermediate work is controlled per conversation with the /verbosity
command, not from this page. See Verbosity.