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

# Chat Interface: Ask Questions About Your AI Activity

> Ask natural-language questions about your AI fleet — Origin queries the analytics graph and returns structured tables and clickable endpoint results.

The **Chat Interface** lets you ask natural-language questions about your endpoints, users, agents, and activity — without writing a query or building a filter. Origin answers by querying the analytics graph, so anything visible in the dashboard is fair game to ask about.

## Opening the Chat

You can launch the chat in two ways:

* Click the **chat icon** (speech bubble) in the top bar of the left panel, next to the date range picker.
* Press **Ctrl+K** (Windows/Linux) or **Cmd+K** (Mac) from anywhere in the console.

## Start Screen

A new chat opens with a prompt and a set of example questions to help you get started. The agent can query all data you have access to across your tenant.

Click any example question to use it as your first message, or type your own question in the **Ask the analytics graph** box. Use **Shift+Enter** to insert a newline without submitting, and type `/` to see available commands.

## Model Selection

Click the **MODEL** indicator in the status bar at the bottom of the chat window to open the model configuration panel.

### Model Tab

The **Model** tab lists available models grouped by status. Each entry shows:

| Column             | Description                                            |
| ------------------ | ------------------------------------------------------ |
| **Context window** | Maximum token capacity (e.g. 1.1M, 2M)                 |
| **Price**          | Cost per million input / output tokens                 |
| **RSN**            | Indicates the model supports reasoning / thinking mode |
| **IMG**            | Indicates the model supports image input               |

Click any model row to apply it to the current and future conversations.

<Note>
  The **TASK MODEL** shown at the bottom of the panel (e.g. `Anthropic: Claude Haiku 4.5 AUTO`) is a smaller model used automatically for internal background tasks such as judging and extraction. It is separate from the main conversation model and does not change when you select a different model for chat.
</Note>

### Provider & API Key Tab

The **Provider & API key** tab configures which AI provider and credentials back the chat.

Select a **Provider** from the dropdown (e.g. Anthropic, OpenAI) and enter your **API key**. The key is stored on the Origin backend and shared across your tenant — the browser only ever sees whether one is configured, never the key itself. Click **Apply** to save.

### MCP Tab

The **MCP** tab connects remote MCP (Model Context Protocol) servers, exposing their tools to the chat agent.

Available integrations include:

| Integration     | Auth    | Description                                                                                             |
| --------------- | ------- | ------------------------------------------------------------------------------------------------------- |
| **Linear**      | OAuth   | Issues, projects, comments, and search                                                                  |
| **Notion**      | OAuth   | Pages, databases, and search across your Notion workspace                                               |
| **Zapier**      | OAuth   | Run any of your Zapier-connected app actions as tools                                                   |
| **Firecrawl**   | API key | Scrape, crawl, search, and map the web — clean markdown returned for LLM context                        |
| **Tavily**      | API key | Search the web with answer-grade snippets purpose-built for LLM agents                                  |
| **Browserbase** | API key | Drive a real headless browser — navigate, click, fill forms, and extract content from JS-rendered pages |

<Note>
  OAuth tokens and API keys for MCP servers are stored in the browser only — they are **not** sent to the Origin backend. Connected integrations show a **CONNECTED** badge and a **Disconnect** button. Tools from connected servers become available to the agent on the next chat message.
</Note>

## Asking a Question

Type your question in the **Ask the analytics graph** box and press **Enter** to submit. The agent reasons over your question and runs one or more tools (e.g. `analytics_query`) against the live dataset. Results are returned as structured tables and/or clickable endpoint chips that link directly to the relevant endpoints in the dashboard.

## Response Actions

Each agent response includes two actions:

* **↓ PNG** — save the response as an image.
* **FORK** — branch the conversation from that point to explore a different follow-up without losing the original thread.

## Conversation Actions

The top-right toolbar of the chat window provides:

* **+** — start a new conversation.
* **↓ PDF** — export the full conversation as a PDF.
* **⤢** — expand the chat to full screen.

## Status Bar

The status bar at the bottom of the chat window shows three indicators:

| Indicator | Meaning                                                      |
| --------- | ------------------------------------------------------------ |
| **READY** | The agent is idle and ready for the next question            |
| **MODEL** | The LLM currently powering the chat; click to change it      |
| **THINK** | Toggle extended reasoning on or off for more complex queries |

## Example Questions

Use these as starting points or adapt them to your own investigation:

* *Give me a list of endpoints and associated users who are actively using Claude Code in the past week*
* *Show me the top 10 tools used across the tenant in the last 7 days, with each tool's share of total tool calls*
* *Plot daily prompt volume for the last 30 days with a 7-day rolling average overlay*
* *Find endpoints whose prompt volume today is ≥1.5× their 7-day baseline mean*
* *Group the last 28 days of prompts by weekday and show me the Mon–Sun pattern*
* *Build me a canvas with the model usage breakdown, latency by provider, and a daily token trend*