Claude advisor
Let Claude Code consult a stronger model at the moments that matter, set per tab from Agentty.
Claude Code can ask a stronger model for advice at key moments in a turn — before a large refactor, when a plan needs judgment, when it is stuck. Agentty exposes that as a setting, so you don't have to remember a flag.
This is an experimental Claude Code capability and it uses extra tokens. It is off unless you choose otherwise.
How it works#
Agentty does not run the advisor itself. When it starts a Claude Code pane it adds --advisor <model> to the command line, and Claude Code does the consulting. That means:
- The advisor uses your Claude Code sign-in and your provider account, like the rest of the session.
- Nothing is written into your configuration files. The choice lives in Agentty's settings and is passed per launch.
- Choosing Off passes a switch that overrides an
advisorModelyou may have set in Claude Code, so "off" really is off for that pane.
Choosing a model#
Settings → General → Claude advisor applies to new Claude tabs:
| Choice | Effect |
|---|---|
| Claude Code setting | Agentty passes nothing; whatever Claude Code is configured with applies |
| Off | The advisor is disabled for that pane, overriding Claude Code's own setting |
| Opus | Consults Opus |
| Fable | Consults Fable |
A pane that is already running can be changed from its status bar, without restarting the session.
When it is worth it#
- Worth it: architectural decisions, a refactor that touches many files, debugging where the first three hypotheses failed, reviewing a plan before the agent commits to it.
- Not worth it: routine edits, tests, formatting, anything where the work is mechanical. The extra tokens buy nothing.
A reasonable default is to leave it off and switch it on from the status bar when a session reaches a decision you care about.