zidaneModels and accounts
Accounts, the model picker, fallback chains, model options, agent profiles and usage in the terminal app.
Which providers exist, how each signs in and how fallback works: providers.
Settings › Authentication#
Settings › Authentication lists each provider and its credential source: the credentials file, OAuth or an environment variable. A provider can hold several OAuth accounts. The footer model chip names the one in use.
| Key | Does |
|---|---|
| Enter | Switch to the provider |
Enter on + add / re-configure | Open the sign-in wizard |
| Tab | Show a provider's accounts |
| CtrlE | Rename an account |
| CtrlX twice | Remove a credential |
Pick a model#
CtrlM opens the model picker with every signed-in provider's models. Type to search by provider, name or capability.
| Key | Does |
|---|---|
| Enter | Use the highlighted model as the primary |
| Tab | Add or remove it as a fallback |
| CtrlShift↑ ↓ | Move it up or down the chain |
| CtrlO | Show or hide older models (on an empty search) |
| Esc | Apply the chain and close (or close, when nothing changed) |
An id the catalog does not list shows a "use anyway" row. The hint row counts selected and matching models.
Build a fallback chain#
The picker lists the chain in order, and zidane tries the first model first. Add fallbacks with Tab and order them with CtrlShift↑ ↓. A chain takes one account per provider, and forks keep it.
A fallback serving a run shows in the footer in the warning color. When a run moves on and back: model fallback.
Set model options#
CtrlL opens the current model's options, when it has any: its reasoning effort levels, then toggles such as fast mode. ↑ ↓ move, Enter or Space picks.
The footer shows the effort and options next to the model: ctrl+m/l Opus 4.6 medium. Effort levels: thinking.
Switch the agent profile#
ShiftTab cycles profiles when you have more than one. The footer shows the current one in its color.
| Profile | Tools |
|---|---|
| Build (default) | All |
| Plan | Read-only |
| Codemode | Build, with tool calls written as one short script. Needs Settings › Agent › Codemode. |
Build and Plan share one agent, so a switch between them keeps the prompt cache: a gate refuses the tools Plan lacks, and a note tells the model about the switch. Plan on a Codemode session works the same way. Other switches rebuild the agent, and the provider reads the conversation again (profiles).
The model, effort and agent pickers mark an option that would re-read a long conversation, for example re-reads ~452k tokens. A switch that keeps the cache shows nothing.
Markdown agents in .agents/agents/ or .zidane/agents/ are sub-agents the main agent starts, not part of this cycle.
Usage#
CtrlP opens the usage panel: rate-limit windows and credits for each OAuth subscription, then GitHub API limits. CtrlR refreshes, Esc closes. Settings › Agent › Check usage stops the fetches. Settings › UI › Usage indicator hides the chip. The windows: plan usage.
Model settings and defaults: Settings. One-off runs take --provider, --model and --thinking: zidane run.