v2026.7.13

Release date: July 18, 2026

---

New Features

Tool: Research information on websites

New ready-made AI Agent action to read web page content during support and use the text (as markdown) as context for the reply.

What it’s for

• Look up catalog, inventory, pricing, or policies on company sites
• Follow changelogs, docs, and product pages without pasting text into the prompt
• Answer with up-to-date web information, limited to the sites you allow

How to use

1. Open the AI Agent (Prompt) and go to Actions / Tools
2. Add the Research information about… action
3. Choose the engine (Firecrawl or Plain HTML), list the sites, and save
4. Publish the prompt and test the conversation

Full guide: AI Tools — Research information on websites

---

Firecrawl integration (organization key)

Register the Firecrawl API key under Settings → Integrations, with key validation and encrypted storage. The key belongs to the organization — it is not stored on the prompt.

What it’s for

• Use the customer’s Firecrawl account (JavaScript / SPA sites)
• Reuse the same key across multiple agents
• Validate the key before saving

How to use

1. Go to Settings → Integrations and find Firecrawl
2. Paste the API key, validate, and save
3. In the AI Agent, select that integration when configuring the tool

Full guide: Firecrawl integration

---

Plain HTML engine (no Firecrawl key)

Besides Firecrawl, the tool can use HTML fetch + markdown conversion, with no integration and no API cost — ideal for static pages.

What it’s for

• Classic HTML sites (little or no JavaScript dependency)
• Environments without a Firecrawl key yet
• Listings and search pages with URL query params (?search=…)

How to use

1. In the action settings, choose the Plain HTML engine
2. Register the allowed sites (URL + description)
3. Save — free web search is disabled in this mode

Limitation

Pages built only with React/Next (content loaded via JS) may return empty or incomplete results with the HTML engine. Prefer Firecrawl in those cases.

---

Site allowlist and subpage navigation

You define the allowed sites (URL + description). The AI can open the site, change search params, and follow subpages under the same path (e.g. /changelog → /changelog/2026/07/), with suggested follow-up links.

What it’s for

• Keep research inside the company allowlist
• Allow catalog search (?search=Civic → ?search=Corolla) without opening the whole web
• Guide the AI when the first page is only an index/menu

How to use

1. Register each site with a URL and a clear description of what is there
2. Optional (Firecrawl only): enable free web search
3. In the conversation, the AI scrapes and, if needed, calls again with follow-up links

---

Free web search (optional)

With the Firecrawl engine, you can allow search on the open web in addition to the site list — useful when the answer may be on public sources.

What it’s for

• Complement the allowlist with open search
• Find public pages before scraping

How to use

1. Configure the Firecrawl integration
2. In the action, enable free web search
3. The AI can use mode=search or scrape public URLs as needed

Security

If free search is off, only URLs from configured sites (and subpages under the same path) are accepted on the server.

---

Benefits

• ✅ The agent answers with up-to-date site content, without pasting pages into the prompt
• ✅ Organization Firecrawl key, encrypted and reusable
• ✅ Plain HTML option for static sites, with no API cost
• ✅ Allowlist and server-side validation — the AI cannot browse where you did not allow