> ## Documentation Index
> Fetch the complete documentation index at: https://docs.praxa.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Core chat and AI pipeline

> Understand the model, reasoning, attachment, and rendered-result capabilities available in Praxa chat.

Understand the model, reasoning, attachment, and rendered-result capabilities available in Praxa chat.

<Note>
  These are Praxa application capabilities. A listed chat capability does not create a public API unless the Capability Atlas links to a developer service.
</Note>

## Provided capabilities

| Capability                                                                                                                                     | Status | Surfaces                   | Who it helps               |
| ---------------------------------------------------------------------------------------------------------------------------------------------- | ------ | -------------------------- | -------------------------- |
| **Multi-model chat**<br />Use one conversation surface across supported frontier models with explicit model selection and thinking controls.   | Live   | iOS, web                   | Individual, Business       |
| **Automatic model routing and cost lanes**<br />Route each turn to an appropriate model lane based on the task, latency, and capability needs. | Live   | iOS, web                   | Individual, Business       |
| **Extended thinking and Adaptive Context**<br />Use visible extended reasoning and bounded contextual preparation for harder requests.         | Live   | iOS, web                   | Individual, Business       |
| **Attachments and vision**<br />Attach images, documents, and files so Praxa can reason over their contents.                                   | Live   | iOS, web                   | Individual, Business, Team |
| **Generative UI and tool cards**<br />Render tool results as task-specific cards instead of exposing raw provider payloads.                    | Live   | iOS, web                   | Individual, Business, Team |
| **Adaptive chat canvas**<br />Present a contextual new-chat home using current personal information when the surface is enabled.               | Live   | iOS, web                   | Individual                 |
| **Runtime capability honesty**<br />Tell the agent exactly which capabilities are available in the current runtime and device context.         | Live   | iOS, web, voice, on-device | Individual, Business, Team |
| **Latency fast lane and prompt caching**<br />Reduce time to a useful answer through warm prompt caches and a bounded fast lane.               | Live   | iOS, web                   | Individual, Business       |
| **Semantic tool selection and reranking**<br />Pilot semantic ranking and health-aware selection over a large tool catalog.                    | Beta   | iOS, web                   | Individual, Business       |

## Example outcomes

1. **Compare two model lanes.** Switch models for the same bounded question and compare the visible response and thinking controls.
2. **Analyze an attachment.** Attach a supported document or image and ask Praxa to identify the evidence it used.
3. **Inspect a tool result.** Ask for current information and confirm the result appears as a purpose-built card rather than raw JSON.
4. **Check runtime honesty.** Ask which capabilities are available in the current runtime and confirm unsupported actions are not presented as available.

## Verify the surface

1. Confirm the capability status and listed surface in the [Capability Atlas](/overview/capabilities).
2. Sign in to the intended Praxa product surface and select the correct personal or organization workspace.
3. Exercise the example with the smallest safe scope and inspect the visible result or approval card.
4. Test one unavailable, denied, or disconnected state so the surface fails honestly.
5. Use a developer tutorial only when this guide links to a supported public developer contract.