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

# Chat copilot

> Use workspace tools with explicit approvals, model controls, and credit limits.

<Frame caption="The workspace assistant.">
  <img src="https://mintcdn.com/auto-gtm/iXt5uMxsbIiTmJfR/images/console-chat.png?fit=max&auto=format&n=iXt5uMxsbIiTmJfR&q=85&s=5109ceacf395619f7b81ecd952e90c4a" alt="The workspace assistant." width="2880" height="1800" data-path="images/console-chat.png" />
</Frame>

Chat answers questions against the active workspace and can prepare or execute supported product actions. Threads, messages, tools, and navigation stay scoped to that workspace.

## Read tools

| Tool                                 | Purpose                             |
| ------------------------------------ | ----------------------------------- |
| <code>get\_dashboard\_summary</code> | Read dashboard scores and summaries |
| <code>get\_timeseries</code>         | Read visibility history             |
| <code>get\_score</code>              | Read the engine scoreboard          |
| <code>list\_runs</code>              | List measurement runs               |
| <code>list\_recommendations</code>   | List recommendation work            |
| <code>get\_report</code>             | Read a report                       |
| <code>list\_prompts</code>           | List prompts                        |
| <code>get\_settings</code>           | Read settings                       |
| <code>list\_connections</code>       | Read connection status              |
| <code>get\_page\_evaluation</code>   | Read a known evaluation             |
| <code>list\_competitors</code>       | List competitors                    |

## Action tools

| Tool                              | Approval required |
| --------------------------------- | ----------------- |
| <code>start\_measurement</code>   | Yes               |
| <code>cancel\_run</code>          | Yes               |
| <code>mark\_recommendation</code> | Yes               |
| <code>add\_prompt</code>          | Yes               |
| <code>edit\_prompt</code>         | Yes               |
| <code>archive\_prompt</code>      | Yes               |
| <code>regenerate\_prompts</code>  | Yes               |
| <code>update\_settings</code>     | Yes               |
| <code>update\_profile</code>      | Yes               |
| <code>set\_credential</code>      | Yes               |
| <code>add\_competitor</code>      | Yes               |
| <code>update\_competitor</code>   | Yes               |
| <code>remove\_competitor</code>   | Yes               |
| <code>set\_model</code>           | Yes               |

<code>draft\_comment</code> prepares text without changing product state. The UI helpers <code>ui\_navigate</code>, <code>ui\_open\_workspace</code>, and <code>ui\_highlight</code> can change the current view but cannot select a workspace the server has not authorized.

## Approval flow

<Steps>
  <Step title="Review the proposed call">
    Chat shows the action and arguments before an approval-gated tool runs.
  </Step>

  <Step title="Approve or reject">
    Approval covers that proposed action. It is not a standing grant for later calls.
  </Step>

  <Step title="Enforce the service contract">
    The underlying route still applies role, scope, lifecycle, and idempotency checks.
  </Step>
</Steps>

## Model switcher

The current allowlist is:

| Display family | Model identifier                         |
| -------------- | ---------------------------------------- |
| Gemini         | <code>google/gemini-2.5-flash</code>     |
| Claude         | <code>anthropic/claude-sonnet-4.5</code> |
| OpenAI         | <code>openai/gpt-5-mini</code>           |
| DeepSeek       | <code>deepseek/deepseek-v4-flash</code>  |

The workspace model is persisted. It overrides the request and environment default. An unrecognized value falls back to Gemini.

## Isolation and redaction

The server constructs tools with a fixed workspace identifier. Threads and messages that belong to another workspace return not found. Credential values are redacted before tool results are written into chat history.

## Metering

Chat authorizes cost for each model step and uses no automatic model-call retry. When the lease cannot authorize another step, the response stops with a credit error before a new provider call.

<Note>
  <code>get\_score</code> currently returns the engine scoreboard. Use the dashboard summary or score routes for the persisted SEO and GEO score cards.
</Note>