Web#

Sero's Web plugin is a built-in plugin (@sero-ai/plugin-web) for bringing web context into an agent session. It can search the web, fetch readable content from sources you provide, retrieve previously stored search/fetch content, run code-oriented web lookups, and keep workspace-scoped bookmarks and activity for later review. See the Plugin Catalog for the built-in plugin inventory.

Sero is still a source-only OSS alpha. Web access depends on configured providers, local runtime support, signed-in browser/profile state for some paths, and third-party services that can change outside Sero.

What the Web plugin does#

The Web plugin focuses on a few practical workflows:

• Search — ask the agent to search the web with web_search or the /web_search slash command.
• Fetch and extract — ask the agent to fetch readable content from one or more sources with fetch_content.
• Retrieve stored content — ask for full content that was previously stored by search or fetch with get_search_content.
• Code search — ask developer-oriented library, API, documentation, or debugging questions with code_search.
• Bookmarks and history — save and manage workspace-scoped bookmarks and search/fetch history with web_bookmark or /web_bookmark.
• Web app and widget — browse recent activity, bookmarks, downloads, and provider status from the Web app or compact Web Activity widget when visible.

These features are provider-mediated. Do not treat Web as a guaranteed plain search engine or universal file extractor.

Search with web_search#

Use web_search when you want current or external information brought into the agent workflow. The tool returns an AI-synthesized answer with source citations. Depending on the provider response and runtime behavior, some source content may also be included or fetched in the background.

Example prompts:

Use web_search to find the current docs for Rspress sidebar configuration.

/web_search search for recent Sero plugin docs and summarize the relevant setup steps

web_search supports multiple related queries. For broader research, it can be useful to ask for a few varied searches instead of one narrow query:

Use web_search with 2-4 varied queries about Electron context isolation best practices.

Search results are added to the session and mirrored into the Web app's workspace state so they can be revisited later.

Fetch and extract content with fetch_content#

Use fetch_content when you already have a URL or source and want the agent to read it directly. The source code includes support paths for pages/URLs, PDFs, GitHub repositories or content, YouTube video transcripts, video-related paths, local video files, and readable markdown extraction from pages.

Those paths are not a promise that every source will extract cleanly. PDFs, GitHub, YouTube/video, and local-video extraction can depend on provider access, runtime dependencies, source availability, and the format of the content itself. Gemini-backed extraction may be used for some fallbacks, so Gemini access can matter for those cases.

Example prompts:

Use fetch_content to read https://example.com/docs/guide and summarize the setup section.

Fetch these three URLs with fetch_content and compare the installation steps.

For a single URL, returned content may be truncated for the chat response. If the full content was stored, use get_search_content to retrieve it by the saved response details.

Retrieve stored content with get_search_content#

get_search_content retrieves content that was previously stored by web_search or fetch_content. It does not fetch arbitrary new URLs on its own. It needs the saved response identifier plus either the query/query index or the URL/URL index from the earlier result.

Use it when the agent says content was stored but the chat response only showed a summary or truncated excerpt:

Use get_search_content to retrieve the full stored content for the second URL from that fetch result.

If there is no stored result for the requested response, ask the agent to run fetch_content again on the source.

Code-oriented web lookup with code_search#

code_search is for developer questions where examples, API docs, snippets, or implementation context are more useful than a broad web answer. It is a limited code-context lookup tool, not a replacement for web_search.

Example prompts:

Use code_search to find examples of Rspress defineConfig sidebar setup.

Use code_search for current TanStack Query mutation examples and summarize the API shape.

Provider behavior for code search should still be treated as alpha. Verify any code examples against the upstream project before applying them.

Bookmarks and history#

Use web_bookmark or /web_bookmark to manage Web plugin bookmarks and related history for the current workspace.

Examples:

/web_bookmark save the first source from the last web search as a bookmark

Use web_bookmark to list my bookmarks for this workspace.

Use web_bookmark to remove the obsolete Rspress migration bookmark.

Bookmarks are workspace-scoped, not global. They are stored with the current workspace's Web state, so a bookmark saved in one workspace should not be assumed to appear in another workspace.

The bookmark tool also manages search/fetch history. Clearing history removes stored search/fetch activity for the workspace and can clear runtime memory for the current session, so use it intentionally.

Web app and Web Activity widget#

The Web app provides a visual place to revisit Web plugin state. The current source-supported surfaces are:

• History — recent searches and fetches
• Bookmarks — saved Web plugin bookmarks
• Downloads — visible downloaded items when present

The Web Activity widget is a compact dashboard summary. It can show counts for searches, fetches, bookmarks, and downloads, provider availability dots, and a small list of recent activity.

Exact app layout and widget placement may change during alpha. Treat the app and widget as browsing surfaces for Web plugin state, while the tools remain the agent-facing way to perform search, fetch, code search, and bookmark actions.

Provider prerequisites#

Sero does not bundle third-party provider credentials. Search and some extraction paths depend on what is configured or signed in for the active profile/session.

Provider paths currently include:

• Exa — can use an API key or an MCP fallback; API-key use may also respect a monthly request budget.
• Perplexity — requires PERPLEXITY_API_KEY or perplexityApiKey in the Web config file.
• Gemini API — requires GEMINI_API_KEY or geminiApiKey in the Web config file.
• Gemini Web — requires signed-in Chromium cookies for a supported Google account in the active profile.

The Web plugin's provider order for web_search is Exa, then Perplexity, then Gemini API, then Gemini Web. Availability can change if credentials expire, provider limits are reached, browser sign-in state changes, or a third-party service changes behavior.

For Gemini Web troubleshooting, the support command google-account can report which Google account appears active. It is a support aid only; it does not make Gemini available by itself.

Where Web state lives#

Web plugin state is stored under the current workspace:

<workspace>/.sero/apps/web/state.json

That state can include search/fetch entries, bookmarks, downloads, provider availability, active provider information, workflow data, and timestamps. Search and fetch results are also appended to session entries and mirrored into this state file.

Because this is workspace state, avoid assuming bookmarks, history, downloads, or provider status are global across every workspace or profile.

Privacy and safety#

Web access can send your query, fetched URLs, and relevant page/source content to the configured provider or model handling the request. Treat web activity as sensitive local workspace state.

Practical safety habits:

• Do not search for or fetch secrets, private tokens, customer data, or internal URLs unless you understand which provider will receive the request.
• Review fetched content before relying on it for security, legal, medical, financial, or production decisions.
• Redact bookmarks, history, URLs, source excerpts, and provider errors before sharing screenshots or support logs.
• Remember that third-party pages can be stale, misleading, hostile, or blocked by their owners.

Troubleshooting and limits#

No provider is available#

Check that at least one provider path is configured: Exa, Perplexity, Gemini API, or Gemini Web. API-backed providers need credentials. Gemini Web needs the right signed-in browser/profile state. Sero does not include default credentials.

Search works but content is missing#

A web_search answer can include citations without full inline source content. Ask the agent to use fetch_content on the cited source, or use get_search_content if the tool stored content for that response.

Fetch returns a short or partial result#

Single-source fetch output can be truncated in chat. Ask for stored full content with get_search_content. If no stored full content exists, rerun fetch_content and narrow the source or section.

PDFs, GitHub, YouTube, or video sources fail#

These are source-supported paths, not guaranteed universal extraction. Failures can come from provider access, format differences, missing transcripts, rate-limits, blocked content, local runtime dependencies, or Gemini availability for fallback extraction.

Bookmarks are missing in another workspace#

Bookmarks live in the current workspace's .sero/apps/web/state.json. Switch back to the workspace where the bookmark was saved, or save a separate bookmark in the new workspace.

Gemini account confusion#

If Gemini Web is expected to work, use the google-account support command to check the active account reported from the profile. Then confirm the profile is signed in and has the cookies needed for Gemini Web-backed behavior.

Provider behavior changed#

Web providers are third-party services. Results, rate limits, auth requirements, content access, and extraction behavior can change without a Sero release. When in doubt, retry with another provider path or fetch the specific source directly.