Skip to main content
Your Digital Twin talks to a real AI model behind the scenes — and which one it uses isn’t a per-message toggle you flip while chatting. The model is part of how the Twin is configured: someone who manages the Twin picks it once, and every conversation uses it until they change it. This page explains how that choice is made, how to tell which model answered a given message, and where to change it if the Twin is yours to manage.
There’s no model switcher floating in the chat box. The model a conversation uses is set per Digital Twin (and, optionally, per assistant) by whoever configures it.

​
Why the model matters

Different jobs deserve different tools, and that’s why Pria supports many models.

Cost

Smaller models burn far fewer credits per turn. For brainstorming or quick lookups, a lightweight model is plenty.

Speed

Compact models reply in seconds. Reserve the slower, larger ones for when you actually need depth.

Depth

Frontier reasoning models (Claude Opus, GPT-5, Gemini 3 Pro, Grok 4) plan, reflect, and self-check before answering — best for analysis, long planning, and nuanced writing.

Specific capability

Some models read images, some have million-token context windows, some power live voice. The right model is the one whose strengths match the task.

​
Which model am I using?

To see exactly which model produced any answer, hover the response and open View Details (the icon). The Conversation History card lists the Model that answered, alongside performance and credit details.
The Conversation History details card showing the model name, performance timings, and credit cost for a single response.
If you simply want to chat, you don’t need to do anything here — your Twin is already set to a sensible model. The rest of this page is for when you manage a Twin and want to choose its model.

​
Setting your Twin’s model

If you manage a Digital Twin, open Settings → Instance → Customize Instance in the sidebar to open its Instance settings, then select the Conversation tab. The Conversation Model dropdown sets the model; expand Model Options to browse the full catalog.
The Instance Settings Conversation tab showing a Conversation Model dropdown set to “-- Default --”, a Model Options expander, and a Max Tokens (Completion) field.
The dropdown’s first choice is labelled -- Default --. Leaving it there means the Twin uses Pria’s current default model — a balanced, reliable choice the Praxis AI team keeps up to date, so you automatically benefit when it’s upgraded. You don’t have to chase model releases yourself.
The change applies to the next question asked. Answers already on screen stay attributed to whichever model produced them.

​
Reading the model catalog

Model Options lists every available model, grouped by provider. Use the Filter providers checkboxes (Amazon Bedrock, OpenAI-CLI, Anthropic Direct API, Google-CLI, Mistral AI, X.AI/Grok — plus Show Deprecated) to narrow the list. Each row shows the model’s Input Size and Output Size (its context window and maximum reply length), an indicative Price $/1M tokens, and a set of Features icons summarising what it can do.
The Model Options catalog with provider filter checkboxes and a table of models from OpenAI, Amazon, and Anthropic showing input size, output size, price per million tokens, and a Features column of small icons.
Pria ships models that comply with Zero Data Retention (ZDR) — your prompts and responses aren’t stored by the model provider. A few models are exceptions, marked with a warning in the catalog; using them accepts the provider’s 30-day retention.
The Features icons tell you at a glance what each model supports:
IconMeaning
Reasoning — supports the reasoning effort scale.
Tools — can call file search, web search, code, and other built-in tools.
MCP connectors — can reach external systems through MCP.
Streaming — replies appear progressively as they’re generated.
Vision — you can paste, drop, or upload images for the model to read.
Structured output — can return clean JSON for downstream use.
Code — strong at generating and explaining code.
1MExtended context — a million-token context window for very large inputs.

​
Custom Models

If your institution has configured its own model — a self-hosted endpoint or one bound to your own provider key — it appears in the model picker as a Custom Model, with its name and endpoint shown above the standard options.
The Settings Instance section showing a Custom Model named gemini-flash at an institution endpoint, with a Max Tokens setting and a Configure Instance link.
Custom Models behave like the built-in catalog from the chat side — same conversation experience, same credit accounting. Setting one up is an institution task; see AI Models in the Admin Guide.

​
Per-assistant model

An assistant (a specialised persona in your Digital Expert Gallery) can run on its own model. When you build or edit an assistant, the Conversation Model field — with the same -- Default -- option and Model Options catalog — lets you pin the model that best fits that persona’s job.
The Edit Assistant dialog with a Conversation Model dropdown set to “-- Default --” and a Model Options expander.
When a conversation runs through an assistant that pins a model, that model is used regardless of the Twin’s default. This is common for image-focused, code, or research assistants where the right model is part of the assistant’s identity. To use a different model with the same prompt, start a conversation with a different assistant, or one that leaves its model on -- Default --.

​
Getting a different model

Open Settings → Instance → Conversation Model and pick the model — or set it on a specific assistant instead, so only that persona changes.
You use the model the Twin is set to. Choose an assistant whose model fits your task, or ask whoever administers the Twin to adjust it. For institution-level model requests, contact the Praxis AI team at humans@praxis-ai.com.