Switching models

Switching models

Ptah lets you change provider and model at any point in a conversation. The selector sits in the chat header — click it, pick a new model, and the next message uses the new one. The existing transcript is preserved in full.

What the selector shows

• The currently active provider (Claude, Copilot, Codex, Gemini, Ollama, OpenRouter, ptah-cli, etc.).
• The currently active model.
• Live availability — providers that aren’t configured or are offline are greyed out with an inline reason.
• A search box for filtering large catalogs (OpenRouter exposes hundreds of models).

What carries over when you switch

State	Carries over?	Notes
Message history	Yes	The full transcript is replayed to the new model.
File attachments	Yes	Stored as part of each message.
Thinking blocks	Provider-dependent	Dropped if the target model doesn’t support them.
Tool results	Yes	Tool-call and tool-result pairs are preserved.
Prompt cache	No	Caches are provider-specific; the new provider starts cold.
Sub-agents already running	Yes	They finish under the provider they were spawned with.

Note

Switching between reasoning and non-reasoning models works, but you may see a quality drop if you move from a reasoning model mid-task to a model without thinking support. Finish the current subtask first when possible.

Switching because of errors

If a provider returns a hard error (rate limit, auth failure, missing model), Ptah surfaces an inline banner with a Switch provider action that opens the selector pre-filtered to healthy options. Your draft message and attachments stay intact.

Per-workspace defaults

Your default provider and model per workspace are stored in ~/.ptah/settings.json:

{
  "llm.defaultProvider": "claude",
  "anthropicProviderId": "openrouter"
}

The in-chat selector overrides these for the current session only. To change defaults permanently, update them from the Providers settings page — see Providers.