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

# July 2026

> Updates and improvements for July 2026

## July 2026

### New features

<AccordionGroup>
  <Accordion title="Official Slack app — one-click install" icon="slack">
    The [Slack integration](/docs/integrations/slack) is now the official SiteGPT app: click **Add to Slack**, pick an escalation channel, and you're connected — no custom app, no manifest, no tokens. The bot answers in every channel you invite it to, on @mentions, and in direct messages, replying in threads with source citations.
  </Accordion>

  <Accordion title="Slack reaction workflows" icon="face-smile">
    React with ❓ (or 🎫 / 🆘) on any message to summon the bot into that thread on demand. React with ✅ to hand the thread to a human — the bot goes silent, the takeover shows in your dashboard, and ❓ brings it back. Escalated conversations cross-post to the channel you chose at install, and the bot introduces itself whenever it joins a channel.
  </Accordion>

  <Accordion title="Slack table rendering" icon="table">
    Tables from your knowledge base now render properly in Slack as aligned grids, switching to a field-by-field layout when a table is too wide for chat.
  </Accordion>

  <Accordion title="SiteGPT plugin on the Crisp Marketplace" icon="comments">
    The [Crisp integration](/docs/integrations/crisp) is now a one-click Marketplace plugin. Install it on your Crisp website, prove ownership with a verification code delivered to your own Crisp inbox, and the bot starts answering visitors — with sources, suggested follow-up buttons, and human takeover (reply as an operator to pause the AI; resolve to hand back). Emails and phone numbers visitors share flow into your SiteGPT leads.
  </Accordion>

  <Accordion title="MCP server: three scoped tools" icon="plug">
    The [SiteGPT MCP server](/docs/cli/mcp-server) now exposes three tools with explicit annotations — `search`, `execute_read` (read-only), and `execute_write` (destructive operations) — so AI assistants like Claude and ChatGPT can group read and write permissions correctly, and structured tool output is available for clients that support it.
  </Accordion>
</AccordionGroup>

### Improvements

* **Model labels**: API v2 responses now report the actual models powering your chatbot — `gpt-4.1` and `gpt-4.1-mini` — instead of legacy tier names. Legacy values remain accepted everywhere as input.
* **Agent sign-in**: agent environments can discover the onboarding flow via `sitegpt.ai/auth.md`, and the CLI ships expanded agent guidance.