Native Search Hub
A browser-driven hub for running search and web-extraction tasks through multiple AI platforms.
Per-dimension scores and reasoning
Evidence shows the repository contains only README and example configs, with no code or scripts, so permissions, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, and source attribution cannot be assessed. All trust criteria scored 0 due to lack of evidence.
Self-consistency: README descriptions of platform integration status (e.g., Gemini marked experimental) are consistent with example outputs, but no code to verify, so scored 1. Dependency availability: No dependencies listed, cannot assess. Failure messages: No error handling or failure message examples provided, scored 0.
Audience and scenarios: README clearly targets OpenClaw and other agent users, and lists various use cases (e.g., travel planning, trend tracking), scored 2. Capability boundaries: Describes supported platforms and search domains, but no explicit limits, scored 1. Trigger precision: Provides example config and CLI usage, but no detailed trigger conditions, scored 1. Environment fit: Mentions support for multiple agent tools, but no specific environment requirements, scored 1.
Information architecture: README is well-structured with sections like install, usage, FAQ, scored 2. Install notes: Only states it's a skill to be installed by agents, but no concrete steps, scored 1. Naming stability: Project and script names consistent in README, but no version info, scored 1. Examples and FAQ: Provides multiple example outputs and FAQ, scored 2. Known limitations: No limitations mentioned, scored 0. License: README marks MIT, but no LICENSE file, scored 1. Versioning/changelog: No version or changelog, scored 0. Maintenance responsibility: No explicit maintainer, but welcomes PRs, scored 1.
Output usability: Example outputs are well-structured and usable for agents, scored 2. Marginal value: Aggregating multiple platform searches reduces redundant work, clear value, scored 2. Cost-benefit: No performance or cost data, but claims reduced maintenance, scored 1.
Claim traceability: Claims (e.g., platform integration status) lack code or test evidence, scored 1. Cross-source corroboration: No external verification, scored 0. Fact-inference separation: No distinction between facts and inferences, scored 0.
- Repository contains only README and example configs, no actual code, so functionality cannot be verified.
- Claims of platform integrations and example outputs may not be validated by actual runs; treat with caution.
- No LICENSE file provided despite README marking MIT.
- No security or privacy notes; browser automation may involve sensitive data.
What does this agent do, and when should you use it?
AI Search Hub is an open-source, browser-driven Skill that brings native search and webpage-understanding capabilities from multiple AI platforms into one entry point. The repository lists Gemini, Grok, Doubao, Yuanbao, LongCat, Tongyi Qianwen, MiniMax, and Kimi as currently connected, while Perplexity, Claude, and ERNIE are presented as future options. Its documented execution entry point is the Python script `scripts/run_web_chat.py`, which takes a site, prompt, and output path and writes results to a local file. The README also includes an `agents/openai.yaml` configuration fragment for agent use and displays support for Claude Code and OpenAI Codex CLI.
python3 scripts/run_web_chat.py accepts a target through --site, sends a query through --prompt, and writes the returned text to the file named by --output. Documented examples use doubao for a Xinjiang travel itinerary and grok for a dated summary of recent X (Twitter) activity with links. The project describes collecting those responses into one output for agents, research processes, and automation workflows; its configuration example explicitly mentions automatic Chrome debug startup and waiting for login.
- A travel-planning workflow that queries Doubao for a Xinjiang itinerary and stores the result as a local text file.
- A researcher tracking live X discussion who uses Grok to organize a public figure’s recent posts by date with links.
- An agent that needs to choose among the documented search sources for global web discovery, Chinese content ecosystems, and social signals.
- An automation pipeline that wants platform-based webpage understanding and extraction for a supplied link instead of maintaining its own page-parsing rules.
- A Chinese market-research workflow seeking supplementary information from WeChat Official Accounts, Douyin, or public webpages through the data ecosystems described for the providers.
What are this agent's strengths and limitations?
- Uses one Python command shape to select a site, submit a prompt, and save the result, making it suitable for scripted workflows.
- Its documented sources span Google web discovery, real-time X social search, Chinese trend content, and WeChat Official Accounts.
- The stated scope includes platform-backed webpage understanding, extraction, cleaning, and organization, rather than only returning search links.
- No verifiable installation, dependency setup, account configuration, or troubleshooting instructions are provided, which makes initial deployment uncertain.
- Core execution depends on browser driving, Chrome debugging, and site login state; reliability is therefore exposed to third-party pages, authentication, and platform availability.
- Future entries such as Perplexity, Claude, ERNIE, and additional vertical sources should not be treated as available capabilities.
- The README describes intelligent routing by question type, but supplies no auditable routing command, interface, or implementation detail that verifies that behavior.
How do you install or deploy this agent?
Installation is not documented. The repository provides no verified post-clone dependency-installation command, Chrome setup, account-login procedure, or credential requirements, so a complete reproducible installation cannot be derived from the supplied material. The explicit runtime evidence is python3, browser-driven operation, and a configuration prompt that refers to Chrome debug startup and login waiting.
How do you use this agent?
The documented first invocation is:
python3 scripts/run_web_chat.py \
--site doubao \--prompt "帮我规划一下新疆旅游路线" \
--output out/doubao_xinjiang_route.txtgrok is another documented site value. A usable browser login state is needed before invocation; the exact initialization, permissions, and login steps are not documented.
How does this agent compare with similar options?
Compared with maintaining custom crawlers, page-parsing rules, and separate browser automation for each platform, AI Search Hub is positioned around reusing connected AI platforms’ native search and webpage-processing entry points and returning consolidated results to an agent or workflow. The trade-off is dependence on those third-party platforms and browser-login flows.