# InteractiveAI Copilot
The Copilot is an AI assistant that operates the InteractiveAI Platform on your behalf. Instead of clicking through the UI or remembering CLI commands and flags, you describe the outcome you want in natural language and the Copilot executes the work, asking for approval before anything is created, updated, or deleted.
The Copilot can be opened in two ways. The chat icon in the platform's right rail opens it as a **side panel** that overlays the page you're on, useful when you want to keep working in the UI and ask the Copilot something in parallel. The Copilot entry in the **sidebar** opens the **full view**, with its own panel listing all your past conversations. Both views share the same conversation history and run under your user identity, scoped to the project you're in.
{% hint style="warning" %}
If you can't see the Copilot entry in the sidebar or the chat icon in the right rail, your plan or your role may not include Copilot access; see the Access section below for details.
{% endhint %}
### Why the Copilot Matters
Operating the platform from the UI or CLI is powerful, but slow. The platform covers a wide range of resources: agents, policies, routines, glossaries, variables, macros, prompts, traces, sessions, services, replicas, and secrets, each with its own commands, fields, and workflows. Even experienced users spend time looking up the exact command, the exact flag, or the exact page they need. The Copilot collapses that lookup time to zero by translating natural-language intent into platform operations:
* **Bridge UI and CLI**: Describe what you want in natural language and the Copilot runs the right commands, parses the results, and renders the output inline.
* **Use real platform knowledge**: It loads the official documentation and component schemas on demand, so it doesn't guess at flags, fields, or workflows.
* **Build, not just read**: It can scaffold an agent, migrate prompts, or wire up a deployment end-to-end, drafting every resource for you to review.
* **Approve every write**: Reads run automatically. Every create, update, or delete pauses for explicit approval before it touches a resource.
* **Audit every step**: Each conversation runs as your user identity and is recorded as a trace you can review, share, or replay.
### What the Copilot Can Do
The Copilot understands a wide range of intents and translates them into platform operations. The most common categories:
| Capability | Example request | Expected behavior |
| ------------------------------- | ---------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Read resources** | "List all my agents", "Show all failed traces from the last 24 hours". | Runs the read command and renders the results as an interactive table or chart. |
| **Create, update, or delete** | "Create a glossary entry for churn rate", "Promote prompt v3 to production" | Drafts the change, asks you to approve it, then executes it. |
| **Analyze traces and sessions** | "Why did this trace fail?", "Group last week's errors by tag" | Pulls the data, summarizes the findings, and links you straight into the trace viewer for deeper inspection. |
| **Build agents end-to-end** | "Help me build a customer support agent" | Asks the right scoping questions, drafts each component (policies, routines, glossary entries, macros), and assembles them into a complete agent configuration. |
| **Deploy services** | "Bump my service to v2.1.0 in dev", "Roll back my-api to the previous image" | Drafts the deployment command, shows what will change, and asks for approval before applying it. |
| **Navigate the platform** | "Open trace 059177", "Take me to the prompts page" | Returns a navigation card that opens the right page in the same window. |
Anything you can do with the platform CLI or UI, the Copilot can do.
### Reads and Writes
The Copilot treats read and write actions differently.
**Reads run automatically.** Listing, fetching, searching, and analyzing never modify anything, so the Copilot executes them and shows you the result inline.
**Writes pause for approval.** Before executing any create, update, or delete, the Copilot opens an approval panel showing:
* The exact command it intends to run.
* The resource it will affect (type, name, project).
* A short, plain-language explanation of the change.
You have three options:
| Action | Effect |
| ----------- | ------------------------------------------------------------------------------------- |
| **Approve** | The command runs and the result is added to the conversation. |
| **Reject** | The command is discarded. The Copilot offers to revise it. |
| **Edit** | You modify the command before it runs (useful for tweaking labels, names, or values). |
{% hint style="info" %}
By default, writes always wait for an explicit decision. This behavior can be overridden via the **Auto-accept write actions** setting in Chat Settings.
{% endhint %}
#### Plan Mode
Plan Mode is enabled from the **+** button next to the prompt input, or from **Chat Settings**. When enabled, the Copilot will not execute any write until it has presented a complete step-by-step plan that you explicitly approve.
**How it works:**
1. You describe the goal, e.g. "Migrate these 2 prompts into the platform and tag them with..."
2. The Copilot may run read-only commands in the background to gather context (existing prompt names, tags, structure).
3. It writes a numbered plan covering every step it intends to take.
4. It asks for a single approval covering the whole plan.
5. On approval, it executes the plan step by step, marking each item as completed.
| Mode | Behavior | When to use |
| ------------- | ------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------- |
| **Default** | The Copilot decides what to do on each turn. Reads run automatically; writes pause for approval. | Day-to-day operations, exploration, single-step tasks. |
| **Plan Mode** | No writes happen until the full plan is approved. | Multi-step migrations, tasks spanning several resources, or any operation you want to review in full before execution. |
{% hint style="info" %}
You can switch modes mid-conversation. In Plan Mode, approving the plan covers all writes inside it; the Copilot will only stop again if the plan changes.
{% endhint %}
### Conversations
Every Copilot conversation is a persistent thread with a title, a project, and a full message history.
* **Conversations panel**: Past conversations are listed in a panel inside the full view, grouped by recency, with a search field at the top. Click any item to reopen it with its full message history, including rendered tables, charts, and diagrams.
* **Project-scoped**: Threads belong to the project where they were created. When you switch projects, the conversations panel updates to show only that project's threads.
* **Auto-titled**: Each thread is automatically titled from your first message. You can rename it at any time, and manual titles are never overwritten.
* **Archive and delete**: Conversations can be archived (hidden from the default list) or permanently deleted from the row menu in the conversations panel.
* **Persistent**: Conversations survive page reloads and browser restarts. You can close the tab mid-task and resume from the same point later.
Long conversations are automatically trimmed to keep responses accurate. Older tool outputs are compacted while the most recent turns are preserved verbatim. The Copilot handles this transparently.
### Rich Outputs
When the Copilot returns structured data, it renders the result as an interactive component rather than as raw JSON or markdown.
| Output | Description |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------- |
| **Table** | Used for lists of resources, query results, and comparisons. Supports sorting, search, and column toggles. |
| **Chart** | Used for distributions, trends, and time-based breakdowns. Available as bar, line, or pie. |
| **Diagram** | Used for routines, workflows, or relationships between components. |
| **JSON tree** | Used for deep, nested structures such as agent configurations, trace inputs, or raw API responses. Nodes are collapsible. |
| **Navigation card** | Opens the relevant platform page directly, such as a specific trace, prompt detail view, or resource list. |
Every rendered output is a regular message in the thread: it scrolls with the conversation, is preserved in history, and is part of the trace.
### Safety and Auditability
* **Every command is visible**: The exact command the Copilot runs is displayed in the message thread alongside its result.
* **Every session is traced**: The Copilot writes its full reasoning and tool calls to **Govern → Traces**, tagged with the project and the user who started the conversation.
* **Reads are limited to your access**: The Copilot operates under your user identity and cannot access resources outside your existing permissions.
* **Writes always require approval**: In default mode, approval is requested per write. In Plan Mode, approval is requested once for the entire plan before execution begins.
### Access
Access to the Copilot is controlled by two independent settings:
| Gate | Scope | Details |
| --------------------- | ------------ | --------------------------------------------------------------------------------------------------------------------------------- |
| **Plan availability** | Organization | Your billing plan must include the Copilot. When it does not, the chat icon and Copilot entry are hidden. |
| **Role capability** | Project | Your project role must include the **Use Copilot** capability. Users without it can view project data but cannot use the Copilot. |
Both conditions are checked on every request. If either is revoked while a session is open, the panel will stop accepting new messages.
If the Copilot is missing or the panel reports an access error, check your role under **Team** settings or contact your organization owner. Plan upgrades are managed under **Organization → Billing.**
### Limits
* **Restricted internet access**: The Copilot can fetch documentation from `docs.interactive.ai` and interact with the platform itself, but cannot access external websites.
* **Tool call timeout**: Individual tool calls time out after 30 seconds to keep conversations responsive. For longer operations such as large migrations or batch updates, use Plan Mode to break the work into smaller approved steps.
* **Step limit per conversation**: Each conversation has a maximum number of automated steps to prevent infinite loops. When the limit is reached, the Copilot pauses and asks whether to continue.
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://docs.interactive.ai/copilot/interactiveai-copilot.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.