Skip to main content
Model choice in Parlé is set at two levels. Your organization decides which models exist at all; each project then narrows that list to the models appropriate for that engagement. Within a project, members pick from what is left, and can switch mid-conversation.

​
Level 1 — The organization’s enabled models

Connecting a provider key does not automatically make every model from that provider available. An Org Admin ticks the models the firm should be able to use in Org settings → Models, and one of them is the organization’s default model for new conversations. The first model you enable becomes the default automatically; use Set default next to any enabled model to change it. There is also an Auto-allow new models toggle per provider — when the provider releases a new model, it is enabled automatically instead of waiting for an admin to opt in. Full detail, including what to do when a model appears greyed out: Connecting Providers.

​
Level 2 — Rules for which models can be used

Every model your organization has enabled is available to every project by default. To narrow that, write a model policy: rules such as “only these providers”, “nothing above this price” or “blocked models”. You set them for the whole organization in Org settings → Models → Model policy, and make them stricter for one project on that project’s Budget tab. Use it to steer a team toward cost-efficient models for routine work, keep an expensive model off a high-volume project, or keep a compliance-sensitive engagement on one provider. Review changes shows what a change would affect before you save it.
1

At project creation

The New project form shows an Available models checklist with every org-enabled model ticked. Untick any you want to block for this project. The section only appears if your organization has models connected.
2

After creation

The project’s blocked models, and any other rules for it, are in its model policy on the Budget tab.
A project’s rules can only narrow the organization’s, never widen them. A model the policy stops stays visible in the picker, greyed out and marked blocked by an admin, so members know why they can’t use it.
If a project’s rules leave no usable model, nobody can chat in that project. Review changes warns you before you save a change that would do this.

​
Let Parlé pick: Auto

On the Business plan and above, the model picker also offers Auto at the top. With Auto, Parlé picks a lower-cost model for routine conversations and a stronger one for demanding ones, from models your Org Admin names, and every answer shows which model wrote it. See Auto, and Set Up Auto for admins.

​
Switching models mid-conversation

You can switch to a different model at any point in an open conversation — mid-thread, mid-task, or after reviewing an initial response. Use the model selector in the chat toolbar and carry on typing. All previous messages — yours and the assistant’s — remain visible and intact, and the new model sees the same thread from the beginning. You can change models partway through a long document review without restarting it.
A productive pattern for billable work: start on a faster, cheaper model for exploration, research and rough drafting, then switch to your most capable model for the final review and polish pass. This keeps spend down without compromising the deliverable.
The model used for each message is recorded with that message and in the token ledger that backs your budgets and usage reporting. You can see exactly which model produced each response — useful for client billing transparency and for reviewing AI-assisted work product.

​
Give the model more time to think

For models that support it, a Thinking control appears beside the model picker. Choose Low, Medium or High to set how much the model reasons before it answers, one message at a time. See Thinking Level.

​
Not every model can use tools

File tools, integrations, and the assistant’s ability to look up your own budgets and policies all depend on the model supporting tool calls. The Available models list in Org settings → Models shows per model whether it supports tools and vision, alongside its context window. If the assistant seems unable to look anything up or reach a file, the model on that conversation is the first thing to check — switch to one that supports tools.
Members of a project can ask “Which models can we use here?” in the conversation itself, and get the list actually available in that project — plus, if one they expected is missing, why. See Asking About Your Org.