Rapidflare Admins
Control web search for your hub
Web search lets your agent use public websites alongside your hub's knowledge to answer a question. Administrators control when it is available, which websites it can access, and how much content it retrieves.
These controls apply to hubs using the Single Front Door (SFD) agent. They do not configure the separate Forge Agent, shown in chat as Pro Agent.
Choose a web search mode
Choose Always On, User Turns On, or Off in the hub's Configure view, under the SFD agent's Web Search settings, then save your changes.
| Mode | What the agent can do | What users see in chat |
|---|---|---|
| Always On | Search public websites when useful. The agent decides whether to search; it does not search for every message. | The Web Search toggle stays on and is disabled. Hover or focus it for an explanation of the hub setting. |
| User Turns On | Search only when the user enables Web Search for the conversation. | The toggle starts off. Users can turn it on or off. |
| Off | Use the hub's other available knowledge and tools without public web search or separate web-page reads. | The Web Search toggle is hidden. Users cannot override the hub setting in chat. |
With User Turns On, enabling the toggle keeps it on for follow-up messages until the user turns it off. It resets when starting or switching conversations, changing hubs, or reloading the page. It is not a preference carried into new conversations.
The toggle is unavailable in shared read-only views and while using Pro Agent. During an answer or attachment upload, users cannot change it.
Restrict which websites the agent can search
Use include_domains to allow specific hosts and exclude_domains to block hosts. Exclusions take precedence, and the same restrictions apply to search results and separate page reads.
Edit these fields in the advanced JSON configuration under agent_config.harness_config.single_front_door.web_search. Merge the settings into the existing configuration rather than replacing unrelated hub settings.
| Entry | Meaning |
|---|---|
"*" | Any public host. Use ["*"] by itself as the include list. |
"example.com" | Only that exact host; it does not include subdomains. |
"*.example.com" | Subdomains of example.com; it does not include example.com itself. |
"docs.example.com" | Only the specified documentation host. |
Enter hostnames, not URLs or paths. For example, use docs.example.com, not https://docs.example.com/guide. To allow an apex domain and its subdomains, list both example.com and *.example.com.
The include list cannot be empty or null, and "*" cannot be combined with other entries. The exclude list may be empty, but cannot contain bare "*"; use Off to disable web search.
For example, this policy enables search automatically but limits it to a documentation host and a support host:
{
"mode": "on_always",
"include_domains": ["docs.example.com", "support.example.com"],
"exclude_domains": [],
"max_results_per_search": 5,
"max_searches_per_turn": 2,
"content": {
"mode": "highlights",
"max_characters_per_page": 3000
},
"max_page_reads_per_turn": 0
}
Replace the example hosts with the websites your team wants the agent to use. Allowlisting a website controls access; it does not certify that every claim on the site is accurate.
Set search and content limits
Search limits control how much evidence the agent can retrieve while answering one user message. They are maximums, not targets the agent must reach.
| Setting | Default | Supported values |
|---|---|---|
mode | on_always | on_always, on_demand, or off |
include_domains | ["*"] | 1–100 host entries |
exclude_domains | [] | 0–100 host entries |
max_results_per_search | 5 | 1–20 results |
max_searches_per_turn | 2 | 1–10 search attempts per user message |
content.mode | highlights | highlights or text |
content.max_characters_per_page | 3000 | 1–50,000 characters |
max_page_reads_per_turn | 0 | 0–10 additional page reads per user message |
Highlights retrieves query-relevant excerpts. Text retrieves page text up to the character limit; it does not guarantee the entire page will be returned. Separate page reads let the agent inspect discovered pages more closely. Setting their limit to 0 disables that additional step, while search results can still include content.
Limits reset for each user message. Failed retrieval attempts also consume the relevant budget. More searches, page reads, and returned content can increase response time and token use. Start with the defaults and adjust them for your team's questions and approved websites.
Recognize information from the web
Web citations link to the public pages used as evidence, so users can inspect the source of a claim. The agent is instructed to place web-derived information under What I found from Web Search: and keep information from organizational content distinct.
This section heading is generated by the agent, so its presentation can vary. Review the citations as well as the answer, especially when public sources and your organization's content disagree. If retrieval is unavailable or produces no usable evidence, the agent may be unable to verify a claim against the web.
Check your configuration
Test the policy with a question that requires information from one of your approved websites, then inspect the answer's web citations.
- For Always On, confirm the toggle is on and disabled; ask a question that benefits from current public information.
- For User Turns On, test with the toggle off and on, then send a follow-up and start a new conversation to check its scope.
- For Off, confirm the toggle is absent and new answers do not retrieve web evidence.
- With domain restrictions, verify that cited websites match the allowed hosts and do not match exclusions.
The current JSON configuration uses web_search.mode. Use that object when editing a hub rather than the retired web_search_enabled boolean.