Skip to main content

​
Overview

Poolside agents include two built-in web tools:
  • web_search: Searches the web and returns ranked results with short excerpts.
  • web_fetch: Fetches a single page and saves its full content for the agent to explore.
Use them when the agent needs information that is not in your workspace or in the model’s training data, such as release notes, changelogs, or current API documentation. web_fetch works by default with no configuration. web_search activates after you configure a search provider with an API key.

​
Configure a search provider

Without a provider, web_fetch retrieves pages with a plain HTTP fetch. Configuring a provider turns on web_search and gives web_fetch cleaner content extraction and excerpts focused on the agent’s objective. Prerequisites You need an API key for one of the supported providers.
Providerprovider value
Parallelparallel
Exaexa
Project settings files such as .poolside/settings.yaml are committed and shared with your team. Keep api_key in a personal file: ~/.config/poolside/settings.yaml or .poolside/settings.local.yaml.
Steps
  1. Add a web_search block to a personal settings file:
    Web search example for ~/.config/poolside/settings.yaml
    web_search:
  provider: parallel
  api_key: <api-key>
  summarize_results: true
  2. Start a new pool session. The agent can now call web_search.
The web_search block supports:
KeyTypeDescription
providerstringSearch provider. Supported values are parallel and exa.
api_keystringProvider API key. Without it, web_search stays off.
base_urlstringOptional override for the provider API endpoint.
summarize_resultsbooleanAdd a summary of results focused on the agent’s stated objective, generated by the session model. Defaults to false.
To disable the built-in web tools, turn off one or both tools:
Disable web tools
tools:
  web_search:
    disabled: true
  web_fetch:
    disabled: true
For more information, see Settings file reference.

​
How the agent uses the tools

For a search, the agent states an objective and optional keyword queries. It can restrict or exclude domains and filter results by publish date. The top results appear inline with excerpts, and the agent saves the full result set to a temporary file so it can page through more results without repeating the search. To read a specific page, the agent calls web_fetch with a URL and an objective. A short excerpt appears inline, and the agent saves the full page content to a temporary file that it explores with its file tools instead of fetching the page again. For URLs the external provider cannot reach, such as localhost, private IP ranges, and single-label intranet hostnames, Poolside fetches the page directly from the agent’s environment instead of sending it to the provider. When summarize_results is true, tool results also include a summary focused on the stated objective. The model that runs your session generates the summary. If summarization fails, the tools return the raw excerpts.